阅读说明：本技术 确定化学生产工厂中的操作条件 (Determining operating conditions in a chemical production plant ) 是由 P·J·霍尔兹迈斯特 F·C·帕特卡斯 C·穆勒 M·弗里德尔 C·M·斯潘格勒 于 2020-02-07 设计创作，主要内容包括：提供了用于确定包括至少一个催化反应器的化学生产工厂的操作条件的系统和方法。经由通信接口接收(10)操作数据和催化剂老化指标。确定(14)用于预定生产运行或当前生产运行的操作条件的至少一个目标操作参数。用于操作条件的至少一个目标操作参数可用于监视和/或控制化学生产工厂。(Systems and methods for determining operating conditions of a chemical production plant including at least one catalytic reactor are provided. Operational data and a catalyst age indicator are received (10) via a communication interface. At least one target operating parameter for an operating condition of a predetermined or current production run is determined (14). At least one target operating parameter for an operating condition may be used to monitor and/or control a chemical production plant.)

1. A system for determining operating conditions of a chemical production plant including at least one catalytic reactor, the system including a communication interface and a processing device in communication with the communication interface,

(a) for a scheduled production run, the system is configured to:

-receiving (10) via the communication interface operation data indicative of predefined operation conditions for the predetermined production run,

-receiving (10), via the communication interface, a catalyst aging indicator associated with a period of time during which the catalyst has been used in the predetermined production run,

-determining (14), via the processing device, at least one target operating parameter for the operating condition of the predetermined production run based on the operating data and the catalyst aging indicator using a data-driven model, wherein the data-driven model is parameterized according to a training data set, wherein the training data set is based on a historical data set comprising operating data, catalyst aging indicators and the at least one target operating parameter,

-providing (16) the at least one target operating parameter for the operating condition of the predetermined production run via the communication interface, or

(b) For a change in a current production run, the system is configured to:

-receiving (10), via the communication interface, measured operation data indicative of current operation conditions for the current production run, wherein at least one operation data point comprises a desired operation value indicative of the change in the current operation conditions,

-receiving (10), via the communication interface, a catalyst aging indicator associated with a period of time during which the catalyst has been used in the current production run,

-determining (14), via the processing device, at least one target operating parameter of the operating conditions for the change in the current production run based on the operating data and the catalyst aging indicator using a data-driven model, wherein the data-driven model is parameterized according to a training dataset, wherein the training dataset is based on a historical dataset comprising operating data, catalyst aging indicator and the at least one target operating parameter,

-providing (16) the at least one target operating parameter for the changed operating condition of the current production run via the communication interface.

2. The system of claim 1, wherein the operational data comprises sensor data measured by a sensor installed in the chemical production plant, quantities derived directly or indirectly from the sensor data, analytical data measured in a sample taken from the chemical production plant, and/or quantities derived directly or indirectly from the analytical data.

3. The system of claim 1 or 2, wherein the system is configured to

-receiving (10) plant metadata indicative of a physical plant layout via the communication interface, an

-determining (14), via the processing unit, at least one target operating parameter additionally based on the plant metadata using a data driven model, wherein the training data set is based on a historical data set additionally comprising plant metadata.

4. The system of claim 3, wherein the system is further configured to preprocess (12), via the processing device, the operational data and the plant metadata prior to determining the at least one target operational parameter, wherein the preprocessing includes a transformation of quantities independent of the physical plant layout.

5. The system of any one of the preceding claims, wherein the historical data set comprises data from multiple runs, multiple plants, and/or multiple catalyst batches.

6. The system of any preceding claim, wherein the catalyst aging indicator is based on a point in time, a period of time, an amount derived from time-dependent operating data, and/or an amount derived from time-dependent operating data accumulation.

7. The system of any one of the preceding claims, wherein the system is configured to determine, via the processing unit, at least one target operating parameter based on determining a short-term model of the target operating parameter at discrete points in time or determining a long-term model of the target operating parameter over a period of time.

8. The system of any one of the preceding claims, wherein the system is configured to receive (10) via the communication interface a time series of the at least one target operating parameter as measured, predicted or derived up to a predicted point in time during the current or previous production run, wherein the system is further configured to determine (14) via the processing unit at least one target operating parameter for one or more points in time after the predicted point in time based on the operational data, the time series of the at least one target operating parameter, the catalyst aging indicator and optionally the plant metadata using the data driven model, wherein the data driven model comprises an inherent time dependency.

9. A system for optimizing operating conditions of a chemical production plant, the system comprising:

the system of any one of claims 1 to 8 and an optimization processing device in communication with the communication interface, the optimization processing device configured to:

-receiving (18), via the communication interface, the determined target operating parameters for more than one operating condition as determined (a) for a predetermined production run or (b) for a change in a current production run,

-determining (18), via the optimization processing device, a minimum or maximum value of a target operating parameter, or a minimum or maximum value of an optimization parameter derived from the target operating parameter, based on the received target operating parameter for each operating condition,

-providing (18) via the communication interface the minimum or maximum value indicative of an optimal operating condition in (a) the predetermined production run or (b) the current production run.

10. A production monitoring and/or control system comprising a communication interface communicatively coupled to the system for determining the operating conditions according to any one of claims 1 to 8 or the system for optimizing the operating conditions according to claim 9.

11. The production monitoring and/or control system of claim 10, comprising a display device configured to receive and display the determined operating conditions; or a control unit configured to receive the determined operating conditions and to control the current production run or the predetermined production run in the chemical production plant based on the determined operating conditions.

12. A computer-implemented method for determining operating conditions of a chemical production plant comprising at least one catalytic reactor, the method comprising the steps of:

(a) for a predetermined production run, the method comprises the steps of:

-receiving (10) via the communication interface operation data indicative of predefined operation conditions for the predetermined production run,

-receiving (10), via the communication interface, a catalyst aging indicator associated with a period of time during which the catalyst has been used in the predetermined production run,

-determining (14), via the processing device, at least one target operating parameter for the operating condition of the predetermined production run based on the operating data and the catalyst aging indicator using a data-driven model, wherein the data-driven model is parameterized according to a training data set, wherein the training data set is based on a historical data set comprising operating data, catalyst aging indicators and the at least one target operating parameter,

-providing (16) the at least one target operating parameter for the operating condition of the predetermined production run via the communication interface, or

(b) For a change in current production operation, the method comprises the steps of:

-receiving (10), via the communication interface, measured operation data indicative of current operation conditions for the current production run, wherein at least one operation data point comprises a desired operation value indicative of the change in the current operation conditions,

-receiving (10), via the communication interface, a catalyst aging indicator associated with a period of time during which the catalyst has been used in the current production run,

-determining (14), via the processing device, at least one target operating parameter of the operating conditions for the change in the current production run based on the operating data and the catalyst aging indicator using a data-driven model, wherein the data-driven model is parameterized according to a training dataset, wherein the training dataset is based on a historical dataset comprising operating data, catalyst aging indicator and the at least one target operating parameter,

-providing (16) the at least one target operating parameter for the changed operating condition of the current production run via the communication interface.

13. The method of claim 12, further comprising the steps of:

-receiving (10) plant metadata indicative of a physical layout of a plant via the communication interface, an

-determining (14), via the processing unit, at least one target operating parameter additionally based on the plant metadata using a data driven model, wherein the training data set is based on a historical data set additionally comprising plant metadata.

14. The method of claim 13, further comprising the step of pre-processing (12) the operational data and the plant metadata via the processing device prior to determining the at least one target operational parameter, wherein the pre-processing comprises a transformation of quantities independent of the physical plant layout.

15. The method of any of claims 12 to 14, wherein the historical data set comprises data from a plurality of runs, a plurality of plants, and/or a plurality of catalyst batches.

16. The method of any of claims 12 to 15, wherein the determination of at least one target operating parameter is based on a short-term model determining the target operating parameter at discrete points in time or a long-term model determining the target operating parameter over a period of time.

17. The method according to any of claims 12 to 16, wherein a time series of the at least one target operating parameter measured, predicted or derived up to a predicted point in time during a current or previous production run is received (10) via the communication interface, wherein at least one target operating parameter for one or more points in time after the predicted point in time is determined (14) via the processing unit based on the operating data, the time series of the at least one target operating parameter, the catalyst aging indicator and optionally the plant metadata using the data driven model, wherein the data driven model comprises an inherent time dependency.

18. A method for optimizing changing operating conditions of a scheduled or current production run of a chemical production plant, the method comprising the steps of:

-receiving (18), via the communication interface, the determined target operating parameters for more than one operating condition as determined (a) for a predetermined production run or (b) for a change in a current production run,

-determining (18), via the optimization processing device, a minimum or maximum value of a target operating parameter, or a minimum or maximum value of an optimization parameter derived from the target operating parameter, based on the received target operating parameter for each operating condition,

-providing (18) via the communication interface the minimum or maximum value indicative of an optimal operating condition in (a) the predetermined production run or (b) the current production run.

19. A method for monitoring and/or controlling a chemical production plant, comprising the step of performing the method for determining operating conditions according to any one of claims 12 to 17 or the method for optimizing operating conditions according to claim 18.

20. A computer program or computer readable non-transitory storage medium comprising computer readable instructions which, when loaded and executed by a processing device, perform the method of any of claims 12 to 19.

21. A method for training a data-driven model for determining a changed operating condition of a scheduled or current production run of a chemical production plant comprising at least one catalytic reactor, the method comprising the steps of:

-receiving a training data set via the communication interface, the training data set being based on a historical data set comprising operation data, catalyst aging indicators, the at least one target operation parameter, optionally plant metadata,

-training, via a processing device, the data-driven model by adjusting a parameterization according to the training data set,

-providing the trained data driven model via a communication interface.

22. A computer program or computer readable non-volatile storage medium comprising the data driven model trained according to claim 21.

23. A catalyst system comprising

A catalyst, and

-a catalyst type identifier associated with a data driven model trained according to the method of claim 21, wherein the model is trained on the catalyst type indicated by the catalyst type identifier.

24. A catalyst system comprising

A catalyst, and

-a catalyst type identifier associated with the computer program or computer readable non-volatile storage medium according to claim 20 or 22.

25. A chemical process using a catalyst associated with a data-driven model trained according to the method of claim 21, wherein the data-driven model is used to design the plant component or optimize the operation of the chemical process for achieving a target performance.

Technical Field

The present disclosure relates to systems, methods, and computer program products for determining operating conditions in a predetermined or current production run of a chemical production plant including at least one catalytic reactor.

Background

Catalytic reactors are widely used in the chemical industry to convert raw materials into valuable chemicals. The properties of the catalyst, such as conversion, selectivity and yield, are related to the operating parameters of the reactor and to the age of the catalyst. To achieve certain performance goals, plant operators must adjust reactor parameters based on experience and engineering knowledge. Typically, reactor models based on kinetic, heat and mass transfer phenomena are used to describe the performance of the catalyst and to achieve more direct control of the reactor operation. However, such models are complex and the experimental determination of kinetic and transport parameters is very cumbersome and expensive.

More recently, mixed-type models have been released that rely in part on process knowledge, but also use machine learning methods. For example, Clough and Ramirez (aichaej 22,1976, p.1097) and Gujarati and Babu (chem. eng.sci.65,2010, p.2009) published knowledge-based models for styrene reactors and used to optimize reactor operation that maximizes styrene yield or process economics. In both cases, the kinetic rate equations and coefficients were taken from previous literature. Considering that the kinetic coefficients should be calculated for each actual catalyst individually (different catalysts have different reaction rates for all chemical conversions occurring in the system), the practical utility of the model is limited. In addition, the catalyst aging rate is not considered.

Tamsiaan et al (comp.chem.40,2012, p.1) developed a model based on kinetic equations disclosed in the literature and fitted the rate coefficients by using styrene plant data from 5 days of operation.

Lim et al (ind. eng. chem. res.43,2004, p.6441) developed a mixed model consisting of: the first principles (knowledge-based) section, which describes the dependence of catalyst performance (conversion and selectivity) on the reactor operating parameters pressure, temperature, STO ratio and ethylbenzene flow (flow rate) as input variables; and a neural network model for predicting inactivation factors to be used in the first principles model. The neural network predicts an updated inactivation factor based on the most recent time step of inactivation factors for ethylbenzene and steam, inlet temperature, total pressure, and partial pressure.

Shahhosseini et al (int.j.chem.fact.eng.9, 2011) developed a hybrid model to optimize the performance of adiabatic industrial reactor systems. First, they fit the kinetic model using 7 data points measured in an experimental isothermal reactor. With the kinetic coefficients estimated in this way, they use tabu search algorithms or genetic algorithms to efficiently optimize the operating conditions to maximize two objective functions: ethylbenzene conversion and styrene selectivity. Based on literature values, catalytic activity was modeled as an exponential function of time, decaying from 100% to 40% over 48 months. Modeling the deactivation in this way does not reflect the impact of different operating scenarios, which is essential for operation and plant specific deactivation prediction.

Wu et al (feature Notes in Computer Science,10357, pp.301-312,2017) and WO 2018/035718A 1 disclose a data-driven model for real-time prediction of styrene production based on real operating data. 33 sensor-based predictors were identified from a combination of Principal Component Analysis (PCA) and random forest variable significance analysis. Three models, Generalized Regression Neural Network (GRNN), kalman filtered GRNN and random forest regression model were compared to predict the yield of styrene monomer. Wu et al focused on predicting current styrene production based on available sensor data, stating that the information is generally only available after analytical testing, which may take several hours. However, the currently proposed methods do not involve performance prediction or performance prediction based on user-defined operating scenarios, including styrene monomer yield at future time points. The deactivation of the catalyst is not taken into account.

Thus, many knowledge-based or kinetic models have been developed to model styrene production in catalytic reactors, and recently elements of Machine Learning (ML) have been integrated (hybrid models), but in some cases are used to optimize the performance of the kinetic model rather than modeling itself. In all hybrid modeling and machine learning approaches, the ML method is trained on a small data set, usually consisting of a single run of a single experimental or production plant.

Knowledge-or kinetic-based mixed reactor models are complex and often require simplifying assumptions in order to reduce computational effort. In most cases, not all physicochemical processes that contribute to catalyst performance and aging are known. For example, in the case of styrene catalysts, potassium loss is controlled by the temperature gradient that occurs along the depth of and within the catalyst bed and the ratio of STO, the carbon dioxide content in the gas mixture (resulting from coke gasification), and the pressure gradient along the bed, making accurate estimation of the potassium vaporization rate impossible. The rate of potassium loss at each point of the reactor bed cannot be calculated a priori in order to estimate the rate of catalyst aging within the catalyst bed and over the life of the catalyst.

It is an object of the present disclosure to provide a method for determining operating conditions in a chemical production plant comprising at least one catalytic reactor, which allows for robust, stable and reliable reactor operation and enhances process control in a catalyst-based production plant.

Disclosure of Invention

According to a first aspect of the present invention, there is provided a system for determining operating conditions of a chemical production plant comprising at least one catalytic reactor. The system comprises:

a communication interface and a processing device in communication with the communication interface, the system configured to:

-receiving, via the communication interface, operation data indicative of predefined operation conditions of a predetermined production run, or measured operation data indicative of current operation conditions of a current production run, wherein at least one operation data point comprises a desired operation value indicative of a change in the current operation conditions,

-receiving, via the communication interface, a catalyst age indicator associated with a period of time for which the catalyst has been used in a current or predetermined production run,

determining, via the processing device, at least one target operating parameter for a predetermined or current production run's varying operating conditions based on the operating data and the catalyst aging indicator using a data-driven model, preferably a data-driven machine learning model, wherein the data-driven model is parameterized according to a training data set, wherein the training data set is based on a historical data set comprising the operating data, the catalyst aging indicator and the at least one target operating parameter,

-providing at least one target operating parameter for a predetermined production run or a changing operating condition of a current production run via the communication interface.

According to another example of the first aspect of the present invention, there is provided a system for determining operating conditions of a chemical production plant comprising at least one catalytic reactor. The system includes a communication interface and a processing device in communication with the communication interface.

(a) For a scheduled production run, the system is configured to:

-receiving, via the communication interface, operation data indicative of predefined operation conditions for a predetermined production run,

-receiving, via the communication interface, a catalyst age indicator associated with a period of time for which the catalyst has been used in a predetermined production run,

determining, via the processing device, at least one target operating parameter for an operating condition of a predetermined production run based on the operating data and the catalyst age indicator using a data-driven model, wherein the data-driven model is parameterized according to a training data set, wherein the training data set is based on a historical data set comprising the operating data, the catalyst age indicator and the at least one target operating parameter,

-providing at least one target operating parameter for an operating condition of a predetermined production run via the communication interface, or

(b) For a change in the current production run, the system is configured to:

-receiving, via the communication interface, measured operation data indicative of a current operation condition of the current production run, wherein the at least one operation data point comprises a desired operation value indicative of a change in the current operation condition,

-receive, via the communication interface, a catalyst age indicator associated with a period of time for which the catalyst has been used in a current production run,

determining, via the processing device, at least one target operating parameter for a varying operating condition of the current production run based on the operating data and the catalyst age indicator using a data-driven model, wherein the data-driven model is parameterized according to a training data set, wherein the training data set is based on a historical data set comprising the operating data, the catalyst age indicator and the at least one target operating parameter,

-providing at least one target operating parameter for a varying operating condition of the current production run via the communication interface.

According to a second aspect of the present invention, a system for optimizing a changing operating condition of a scheduled or current production run of a chemical production plant is provided. The system comprises the system and an optimization processing device, wherein the optimization processing device is configured to:

-receiving the determined target operating parameter via the communication interface for more than one operating condition of the predetermined production run or of a change of the current production run,

-determining, via the optimization processing device, a minimum or maximum value of the target operating parameter, or of at least one optimization parameter derived from the target operating parameter, based on the received target operating parameter for each operating condition,

-providing via the communication interface a minimum or maximum value indicative of an optimal operating condition of the predetermined or current production run, e.g. a minimum or maximum value of the target operating parameter, or a minimum or maximum value of an optimization parameter derived from the target operating parameter.

According to another example of the second aspect of the present invention, there is provided a system for optimizing operating conditions of a chemical production plant. The system comprises the system and an optimization processing device, wherein the optimization processing device is configured to:

receiving the determined target operating parameters via the communication interface for more than one operating condition as determined (a) for a predetermined production run or (b) for a change in a current production run,

-determining, via the optimization processing device, a minimum or maximum value of the target operating parameter, or a minimum or maximum value of an optimization parameter derived from the target operating parameter, based on the received target operating parameter for each operating condition,

-providing via the communication interface a minimum or maximum value indicative of an optimal operating condition in (a) a predetermined production run or (b) a current production run, e.g. a minimum or maximum value of a target operating parameter, or a minimum or maximum value of an optimization parameter derived from the target operating parameter.

According to a third aspect of the present invention, there is provided a production monitoring and/or control system comprising a communication interface communicatively coupled to a system for determining or a system for optimizing operating conditions as listed above, e.g. via a wired or wireless connection. The production monitoring and/or control system may include a display device configured to receive and display the determined operating conditions. The production monitoring and/or control system may include a control unit configured to receive the determined operating conditions to control a current or predetermined production run in the chemical production plant based on the determined operating conditions. The determined operating conditions preferably include the determined target operating parameters and optionally further operating data including desired operating values.

According to a fourth aspect of the present invention, there is provided a computer-implemented method for determining a changed operating condition of a scheduled or current production run of a chemical production plant comprising at least one catalytic reactor. The method comprises the following steps:

-receiving, via the communication interface, operation data indicative of predefined operation conditions of a predetermined production run, or measured operation data indicative of current operation conditions, wherein at least one operation data point comprises a desired operation value indicative of a change in the current operation conditions,

-receiving, via the communication interface, a catalyst age indicator associated with a period of time for which the catalyst has been used in a current or predetermined production run,

determining, via the processing device, at least one target operating parameter for a changing operating condition of a predetermined or current production run based on the operating data and the catalyst aging indicator using a data-driven machine learning model, wherein the data-driven model is parameterized according to a training dataset, wherein the training dataset is based on a historical dataset comprising the operating data, the catalyst aging indicator and the at least one target operating parameter,

-providing at least one target operating parameter for a predetermined or changing operating condition of the current production run via the communication interface.

According to another example of the fourth aspect of the present invention, there is provided a computer-implemented method for determining operating conditions of a chemical production plant comprising at least one catalytic reactor. The method comprises the following steps:

(a) for a scheduled production run, the method comprises the steps of:

-receiving, via the communication interface, operation data indicative of predefined operation conditions for a predetermined production run,

-receiving, via the communication interface, a catalyst age indicator associated with a period of time for which the catalyst has been used in a predetermined production run,

determining, via the processing device, at least one target operating parameter for an operating condition of a predetermined production run based on the operating data and the catalyst age indicator using a data-driven model, wherein the data-driven model is parameterized according to a training data set, wherein the training data set is based on a historical data set comprising the operating data, the catalyst age indicator and the at least one target operating parameter,

-providing at least one target operating parameter for an operating condition of a predetermined production run via the communication interface, or

(b) For a change in current production operation, the method comprises the steps of:

-receiving, via the communication interface, measured operation data indicative of a current operation condition of the current production run, wherein the at least one operation data point comprises a desired operation value indicative of a change in the current operation condition,

-receive, via the communication interface, a catalyst age indicator associated with a period of time for which the catalyst has been used in a current production run,

determining, via the processing device, at least one target operating parameter for a varying operating condition of the current production run based on the operating data and the catalyst age indicator using a data-driven model, wherein the data-driven model is parameterized according to a training data set, wherein the training data set is based on a historical data set comprising the operating data, the catalyst age indicator and the at least one target operating parameter,

-providing at least one target operating parameter for a varying operating condition of the current production run via the communication interface.

According to a fifth aspect of the present invention, a method for optimizing a changing operating condition of a scheduled or current production run of a chemical production plant is provided. The method comprises the following steps:

-receiving the determined target operating parameter via the communication interface for more than one operating condition of the predetermined production run or of a change of the current production run,

-determining, via the optimization processing unit, a minimum or maximum value of the target operating parameter, or a minimum or maximum value of an optimization parameter derived from the target operating parameter, based on the received target operating parameter for each operating condition,

-providing via the communication interface a minimum or maximum value indicative of an optimal operating condition of the predetermined or current production run, e.g. a minimum or maximum value of the target operating parameter, or a minimum or maximum value of an optimization parameter derived from the target operating parameter.

According to another example of the fifth aspect of the present invention, a method for optimizing a changing operating condition of a scheduled or current production run of a chemical production plant, the method comprising the steps of:

receiving the determined target operating parameters via the communication interface for more than one operating condition as determined (a) for a predetermined production run or (b) for a change in a current production run,

-determining, via the optimization processing device, a minimum or maximum value of the target operating parameter, or a minimum or maximum value of an optimization parameter derived from the target operating parameter, based on the received target operating parameter for each operating condition,

-providing via the communication interface a minimum or maximum value indicative of an optimal operating condition in (a) a predetermined production run or (b) a current production run, e.g. a minimum or maximum value of a target operating parameter, or a minimum or maximum value of an optimization parameter derived from the target operating parameter.

According to a sixth aspect of the present invention, there is provided a method for monitoring and/or controlling a chemical production plant comprising the steps of performing the method for determining or optimizing the operating conditions as listed above. The method may further include displaying the determined operating conditions on a display device and/or controlling a current or predetermined production run in the chemical production plant based on the determined operating conditions. The determined operating conditions preferably comprise the determined target operating parameters and/or optionally operating data comprising desired operating values.

According to a seventh aspect of the present invention, there is provided a method for training a data-driven model for determining a changed operating condition of a scheduled or current production run of a chemical production plant comprising at least one catalytic reactor. The method comprises the following steps:

-receiving a training data set via the communication interface, the training data set being based on a historical data set comprising operation data, catalyst aging indicators, at least one target operation parameter, optionally plant metadata,

training, via a processing device, the data-driven model by adjusting the parameterization according to a training data set,

-providing the trained data driven model via a communication interface.

According to an eighth aspect of the invention, a computer program or a computer program product or a computer readable non-volatile storage medium comprising computer readable instructions which, when loaded and executed by a processing device, performs the method disclosed herein.

According to a ninth aspect of the present invention there is provided a catalyst comprising a catalyst type identifier associated with a data driven model trained in accordance with the method set out herein, wherein the model is trained on a catalyst type indicated by the catalyst type identifier. In other words, a catalyst system is provided that includes a catalyst and a catalyst type identifier associated with a data-driven model trained according to the methods set forth herein, wherein the model is trained on the catalyst type indicated by the catalyst type identifier.

According to a tenth aspect of the present invention there is provided a catalyst comprising a catalyst type identifier associated with a computer program as set out herein. In other words, a catalyst system is provided that includes a catalyst and a catalyst type identifier associated with a computer program listed herein.

According to an eleventh aspect of the present invention, there is provided a chemical process using a catalyst associated with a data-driven model trained in accordance with the method set forth herein, wherein the data-driven model is used to design plant components and optimize the operation of the chemical process for achieving a target performance, in particular the operating conditions of a chemical production plant comprising at least one catalytic reactor.

The following disclosure applies to, among other things, the systems, methods, computer programs, computer-readable non-volatile storage media, catalysts, chemical processes, and computer program products disclosed herein. Accordingly, no distinction is made between systems, methods, computer programs, computer-readable non-volatile storage media, or computer program products. All features are disclosed in relation to the systems, methods, computer programs, computer readable non-volatile storage media, catalysts, chemical processes, and computer program products disclosed herein.

The present invention provides a system or method for determining operating conditions or operating parameters based solely on data-driven models that allows for more robust, stable, and reliable reactor operation and enhances process control in a catalyst-based production plant. In particular, the determination is more accurate than known methods based on kinetic or mixed models, since no a priori information about the reaction kinetics is required, and thus no estimation or simplifying assumptions about the underlying process are required.

Specifically, the determination takes into account catalyst deactivation by providing an indication of catalyst aging associated with the period of time that the catalyst has been used in the current production run. The inclusion of a catalyst aging indicator as a model input parameter allows for more accurate determination of operating conditions because the model inherently takes into account catalyst deactivation or aging. The determination of operating parameters is applicable to a wide range of chemical production plants and operating parameters, taking into account catalyst deactivation. In contrast, data-driven models that do not account for catalyst aging are only narrowly applicable. In particular, such models are limited to the chemical production plant for which the model is trained, and sometimes even run only once in the production plant, the range of operating conditions is narrow.

The system and method of the present disclosure performs short-term discrete prediction as well as time series prediction. In the latter case, they are able to cover the complete deactivation process of the catalyst during the entire production run. The method and system enable plant operators to refine and optimize operating strategies on a daily basis based on energy costs, market supply of raw materials or demand for plant products, and other constraints that may occur in the plant (e.g., blackouts in different plant sections or utilities).

Further, the system and method allow for the use of plant operational data and catalyst aging indicators to predict the short term behavior and/or forecast the long term behavior of a catalyst based production plant. This allows for enhanced process control in a catalyst-based production plant, since, for example, plant operators can easily assess the intended or current operation based on the operating conditions present in the production plant. Thus, the system provides a powerful tool for planning, monitoring and controlling a production process.

In one embodiment of the invention, a chemical production plant includes one or more catalytic reactors. In the context of the present invention, the term "catalytic reactor" refers to a chemical reactor in which a catalytic chemical reaction takes place and which typically comprises a catalyst. The catalytic reactor may be a fixed bed catalytic reactor. The chemical production plant may be a styrene production plant.

In this context, a data-driven model, preferably a data-driven machine learning model or a data-driven model only, refers to a trained mathematical model parameterized according to a training data set to reflect the reaction kinetics or physicochemical processes of a chemical production plant or catalytic reactor. Untrained mathematical models refer to models that do not reflect reaction kinetics or physicochemical processes, e.g., untrained mathematical models do not stem from the laws of physics that provide scientific generalizations based on empirical observations. Thus, the kinetic or physicochemical properties may not be inherent to the untrained mathematical model. Untrained models do not reflect this property. The parameterization of the untrained mathematical model can be realized by performing feature engineering and training by adopting a corresponding training data set. The result of such training is a data-driven model only, preferably a data-driven machine learning model, which reflects the reaction kinetics or physicochemical properties as a result of the training process, preferably as a result of the training process only.

Historical data in this context refers to data sets that include at least operating data, catalyst aging indicators, and at least one target operating parameter, where each data set is associated with a single production run. Thus, each data set includes data associated with a production run of one catalyst from the beginning of production to the end of production. Such data can be measured and recorded during a production run over the life of the catalyst (e.g., from the beginning of production to the end of production after catalyst replacement) when the catalyst needs to be replaced again.

The chemical reaction catalyzed does not limit the scope of the invention. By way of example, the catalyst may be a dehydrogenation catalyst. In particular, the catalyst of the catalytic reactor may be an iron oxide based catalyst for the dehydrogenation of aliphatic or alkylaromatic hydrocarbons to form the corresponding unsaturated hydrocarbons. Examples of such dehydrogenation processes are the dehydrogenation of ethylbenzene to styrene, cumene to alpha-methylstyrene, butene to butadiene or isoamylene to isoprene. The method and system are particularly useful in styrene production plants. Preferably, the chemical production plant is a styrene production plant that converts ethylbenzene to styrene using a styrene catalyst. The production of styrene generally involves heterogeneously catalyzed dehydrogenation of ethylbenzene in the presence of steam. The catalytic dehydrogenation of aliphatic or alkylaromatics is generally carried out industrially in the presence of steam at temperatures in the range from 500 to 700 ℃. In these processes, the hydrocarbon and steam are typically mixed at high temperature and low pressure and passed over an iron oxide dehydrogenation catalyst.

The term operational data refers to quantities indicative of the operational state of a chemical production plant. In particular, such quantities relate to measurement data collected during a production run of the chemical production plant and may be derived directly or indirectly from such measurement data. In a preferred embodiment, the operational data comprises sensor data measured by a sensor installed in the chemical production plant, quantities derived directly or indirectly from such sensor data, analytical data measured in a sample taken from the chemical production plant, quantities derived directly or indirectly from such analytical data, or combinations thereof.

The sensor data may include measured quantities available in the chemical production plant by way of installed sensors (e.g., temperature sensors, pressure sensors, flow sensors, etc.). The analytical data may include the amount provided by analytical measurements of samples taken from a chemical production plant at any point in the process or time. In particular, such analytical data may include the composition of reactants, starting materials, products and/or by-products as determined, for example, via gas chromatography from samples extracted during the production process at different stages of the production process (e.g., before or after a catalytic reactor). The analytical data preferably form the basis for determining the performance characteristics of the catalyst.

The operational data set may include raw data, which refers to basic, raw analytical and/or sensor data; or processed or derived parameters derived directly or indirectly from the raw data. In the case of a chemical production plant, the derived parameter may comprise an average inlet temperature of the plurality of catalytic reactors derived from the respective temperature sensors; a steam-to-oil ratio derived from raw data of steam flow rate and reactant flow rate; conversion and selectivity derived from analytical data before and after the reactor; any type of normalized data, such as production values normalized by catalyst volume or catalyst mass; any data derived from the time series data, such as cumulative throughput, maximum load to date, or any combination thereof.

Especially for chemical production plants, conversion, selectivity and yield can be derived from the analytical data. Conversion here means the fraction, preferably the percentage, of reactants that are completely converted in the reactor. In the case of styrene production, for example, this corresponds to the conversion of the starting material ethylbenzene to any product. Selectivity to the desired product refers to the amount of conversion reactant that is converted to the desired product. In the case of styrene production, for example, this corresponds to the selectivity of the ethylbenzene to styrene reaction. The yield of the desired product refers to the mathematical product of the conversion and the product-specific selectivity. The yield can be expressed as the percentage of reactants entering the reactor that are converted to the desired product. In another embodiment, plant metadata indicative of a physical plant layout is received via a communication interface. The plant metadata may include plant-specific quantities describing characteristics of, for example, the reactors, which are predefined by the physical plant layout and may be related to the performance of the plant or reactors. For example, plant metadata includes the number of reactors through which the reactant mixture subsequently passes, e.g., 2 or 3 reactors, total catalyst volume, catalyst volume of a reactor, size of each reactor (length, diameter, height … …), type of catalyst used in the plant, or a combination thereof. In another embodiment, determining, via the processing unit, the at least one target operating parameter is based additionally on plant metadata using the data driven model, wherein the training data set is based on a historical data set additionally including the plant metadata. The determination of incorporating plant metadata into operating conditions allows for the construction of data driven models that are applicable to different plants, which in turn increases the number of data points available for training the data driven models. Thus, the data driven model broadly captures the operating conditions of different plants operating under different operating conditions with different physical plant layouts, which allows for a more accurate determination.

In another embodiment, the set of historical data includes data from multiple production runs, multiple plants, and/or multiple catalyst batches of the same type of catalyst. Incorporating multiple production runs into the training allows for different operating conditions to be covered for the same or different plants. Including data from multiple catalyst batches allows for accounting for differences between catalyst batches. The data from multiple catalyst batches may include data from multiple production runs, wherein for at least one production run, different production batches of the same type of catalyst are used. This may be included by including multiple plant data from one or more production runs of different plants. Thus, operating conditions in different plants may be covered, providing a wider applicability of the model. In this context, the same type of catalyst refers to the same type of catalyst formulation. Multiple catalyst lots include the same type of catalyst provided from different manufacturing lots or different delivery dates.

In another embodiment, the catalyst age indicator is based on a point in time, a period of time, an amount derived from time-dependent operating data, and/or an amount cumulatively derived from time-dependent operating data. The catalyst age indicator may be specified by the time the catalyst spends in the reactor under the reaction conditions since it was first contacted with the reaction mixture. Additionally or alternatively, the cumulative loading or cumulative throughput of the catalyst may be used as an indicator of catalyst aging, which is preferably defined by the total amount of reactant feed or converted reactant from the start of the run up to the predicted start time point. In addition to the indicators mentioned herein, any other amount useful as an indicator of catalyst aging may be utilized. The catalyst aging indicator may be provided via a client device, wherein a plant operator inputs a time at which a production run begins or a time period from the beginning of the production run until a predicted start time point. Alternatively or additionally, the catalyst aging indicator may be determined based on a time series of operational data, preferably from the start of a production run, where the start of a production run may be determined based on an operational profile in the operational data. Such an operating profile may include certain temperature, pressure, flow profiles, or a combination of such profiles.

Operating conditions for a scheduled or current production change refer to operating conditions under which the chemical production plant may be operating in the future or after a predicted start point in time. Such operating conditions may include one or more operating conditions at one or more discrete points in time, at a plurality of discrete points in time, over a period of time, or for several time intervals. In the latter case, the operating conditions may comprise different time intervals for which the at least one predefined or desired operating value takes different values in different intervals. A plurality of discrete points, periods or intervals of time may extend from the predicted start point of time to the remaining production runs until the end of the production run, at which time the catalyst needs to be replaced.

The operating conditions for a predetermined production run may refer to the operating data of predefined operating conditions of the chemical production plant prior to the start of the production run. The predefined operating conditions may include a set of predefined operating data specifying the operating conditions. Determining operating conditions for a predetermined production run is particularly useful for production design and planning prior to the start of the production run. The operating conditions for the current production run may refer to changes in the operating conditions as currently set in the chemical production plant. The operating conditions may include a set of operating data specifying current operating conditions, wherein the operating data is located in at least one operating data point that is different from an operating parameter currently set in the chemical production plant. Determining operating conditions for a current production run is particularly useful for monitoring and controlling production during the current production run.

Determining operating conditions includes predicting or forecasting performance/behavior of the catalyst or reactor and is used in a general manner to describe the application of a data-driven model to a set of appropriate input parameters. In the example of a styrene production plant, for example, this determination refers to a determination of the reactor inlet temperature and the associated selectivity of the ethylbenzene to styrene reaction given specific operating conditions via the operating data. This selection of the output parameters of the model may be driven by the fact that: most plants are operated in a manner where an operator adjusts the inlet temperature to achieve a determined desired conversion. However, given the same data set, the model may be developed in a similar manner, as described in further detail below, with different input and output allocations, for example using reactor temperature as an input to predict the expected conversion at operating conditions.

In one embodiment, the determination of the at least one target operating parameter for the operating condition is based on a short-term model determining the target operating parameter at discrete points in time or on a long-term model determining the target operating parameter over a period of time, in particular in the future. In the case of a short-term model, at least one target operating parameter for the operating condition is determined based on a short-term framework. Here, a short time frame may refer to a discrete or single point in time. The short-term model may be based on different machine learning techniques including, for example, regression models such as linear regression models, non-linear regression models, bayesian linear regression, random forest models, neural networks, or combinations thereof. Other methods may also be applied. Preferably, the short-term model has no inherent time dependence. More preferably, the short-term model predicts the performance or operating condition at time t based on the model inputs at the same time t.

In the case of a predetermined production run, the operational data may specify predefined operating conditions. The catalyst aging indicator may be estimated for a predetermined production run, for example, under the assumption that predefined operating conditions are constant over time or have predefined changes over time or based on previous production runs. The determination via the short-term model may for example be applied to more than one discrete point in time for maintaining constant predefined operating conditions throughout the production run, and the catalyst aging indicator may be estimated for each of the discrete points in time. This implementation is particularly advantageous for designing and planning future production runs.

In the case of a current production run, the operational data may specify operating conditions currently set in the chemical production, wherein the at least one desired operating value indicates a change in or deviation from the current operating conditions. The determination via the short-term model may, for example, be applied to one or more discrete points in time for determining the effect of at least one desired operating value indicative of a deviation from an operating condition currently set in chemical production. This implementation is particularly advantageous for monitoring and controlling current production runs.

In the case of the long-term model, at least one target operating parameter for the operating condition is determined based on the long-term framework. The long time frame herein may refer to a plurality of time points in the future. The number of points and thus the prediction horizon depends on the time scale of the time dynamics in the production plant. For catalyst-based production processes, this dynamics can be determined by the catalyst aging dynamics and their time frame. For heterogeneous catalyst reactions like styrene production, such time scales may be in the range of weeks, months or years. The long-term model may be based on a time series forecasting method. Such methods include, for example, known regression methods such as autoregressive models, in particular Autoregressive (AR), Moving Average (MA), autoregressive moving average (ARMA), autoregressive integrated moving average (ARIMA), Vector Autoregressive (VAR), vector autoregressive moving average (VARMA), vector autoregressive moving average with exogenous regressions (VARMAX), random forest models, neural networks, convolutional neural networks, recursive neural networks, or combinations thereof. Other methods may also be applied.

In one embodiment, a time series of at least one target operating parameter as measured, predicted, or derived during a current or previous production run up to a predicted start time point is received via a communication interface. For example. In the case of scheduled operations, target operating parameters as measured, predicted, or derived during previous production may be used. Preferably, the at least one target operating parameter for one or more points in time after the predicted start point in time is determined using a data-driven model based on the operational data optionally including the desired operational value, the time series of the at least one target operating parameter, the catalyst aging indicator, and optionally the plant metadata. More preferably, the data-driven model includes an inherent time dependence. In this embodiment, the at least one target operating parameter may comprise an uncontrolled or endogenous parameter that is not controllable via machine settings in the chemical production plant. Rather, the operational data including at least one desired operational value may include a controlled or exogenous parameter that is controllable via machine settings in the chemical production plant. The determination of the target operating parameters at the predicted start time t may include, for example, a determination for time t, …, t + N, N >0 based on uncontrolled parameters or a subset of uncontrolled parameters up to a time point t-1 or less, and optionally based on controlled parameters or a subset of controlled parameters, such as at time point t, …, t + N (optionally further including, for example, t-1, t-2.,) or at other time points suitable for forecasting using the particular structure of the selected model.

Preferably, the long-term model is a time-series based model that includes inherent time dependencies and forecasts the target operating parameters at time t, …, t + N based on model inputs up to time point t-1. Here t refers to the starting point of the prediction in time. Thus, the long-term model allows for determination of reactor or catalyst performance and changes in reactor or catalyst performance over a preferably extended period of time without the available information of uncontrolled parameters exceeding the forecast starting point. These models have at least some inherent temporal correlation and predict the performance of time t, …, t + N based on uncontrolled parameters up to time t-1 and optionally on controlled parameters (e.g., at time t, …, t + N or other points in time suitable for prediction using the particular structure of the selected model).

In one embodiment, the processing device is further configured to pre-process the operational data prior to the determination via the data driven model. Preferably, the preprocessing includes conversion of quantities independent of the physical plant layout. The pre-processing allows for systematic and non-systematic differences between different plants to be taken into account. Thus, even if the data from the particular plant in question is not used to train the data-driven model, it can be widely applied to the determination of target operating parameters. In particular, the transformation includes system factors as input parameters for data-driven model and/or operational data normalization.

In another embodiment, the data-driven model is validated prior to determining the at least one target operating parameter. Such verification enhances the interpretability and trustworthiness of the determined target operating parameter. For verification, operational data and at least one target operational parameter as measured or derived for one or more points in time during a current production run may be received via the communication interface. The operational data may be used to determine at least one target operational parameter corresponding to the point in time, where the measured target operational parameter is available. The results of the determined target operating parameter for a certain point in time and the measured operating parameter corresponding to the same point in time may be compared. If the comparison results in valid model operation, for example if the difference is less than a threshold or no systematic error is identified, then a determination of at least one target operating parameter for the operating condition based on a set of operating data including at least one desired operating value may be followed. If the comparison results in an invalid model operation, an alarm may be triggered, for example via a display device or an audio device, indicating that the model operation is not suitable for monitoring and/or controlling the target chemical plant under the current operating conditions.

In another embodiment, the data-driven model is selected based on the catalyst type prior to determining the at least one target operating parameter. The catalyst type may be received via metadata that indicates the type of catalyst used, for example via a catalyst type identifier. The catalyst type may specify the catalyst formulation. This option allows a high degree of flexibility in the use of the system, computer program product and method, covering not only different chemical production plants, but also different catalyst types. A further degree of flexibility may be added by selecting the data-driven model based on the input and output parameter identifiers prior to determining the at least one target operating parameter. The input and output parameter identifiers may specify which parameters are used as operational data and which parameters are used as target operational parameters in the system or method. In such implementations, different data-driven models may be stored in the memory of the system, including catalyst types and/or parameter identifiers for each model, based on the selection steps that may be performed.

In one embodiment, training of the data-driven model is performed based on training data that includes a catalyst type identifier. The trained models may be stored with the catalyst type identifier and may be provided with a catalyst of the type that preferably includes the catalyst type identifier. Such identification may be accomplished electronically via a catalyst ID stored in a database or, for example, in conjunction with a mobile storage medium attached to the catalyst delivery vessel at the corresponding chemical production plant. In this case, the model is preferably trained based on a training data set measured in a production run using the same catalyst as the catalyst type indicated by the catalyst type identifier. Thus, the catalyst may be bundled with a data-driven model and may help provide more robust control of a production plant running with a catalyst of a particular catalyst type.

A method is provided for predicting the short term behavior of a catalyst and predicting the long term behavior of a catalyst, for example in a fixed bed catalytic reactor, as a function of reactor operating parameters or operating data and an indicator of catalyst aging or catalyst aging. The method uses a data-driven model, preferably a data-driven machine learning model, which does not involve prior information about reaction kinetics. The model is capable of predicting both short-term and long-term behavior of the catalyst as a function of input parameters, including typical reactor operating parameters or operating data, as well as parameters derived from sensors and analytical raw data available in the production plant. A software product for performing the method is also provided. As an example of an application, the method is used to predict and predict the behavior of a catalyst and process reactor for the conversion of ethylbenzene to styrene.

The present disclosure provides a computer-implemented method of predicting short-term performance and forecasting long-term performance (including catalyst aging effects) of a catalyst in a chemical production plant including a catalytic reactor. The method involves using a mathematical model of the chemical production plant and in particular of the catalytic reactor, which is based on machine learning, does not involve prior information of the reaction kinetics, and uses input parameters selected from the group consisting of sensor raw data, derived parameters, reactor operating parameters or operating data, plant metadata and parameters indicative of catalyst aging or catalyst aging indicators.

In one embodiment, the operational data is selected from sensor data obtainable from chemical production plants, in particular catalytic reactors, analytical data from, for example, Gas Chromatography (GC) analysis, and derived parameters as listed above.

In one embodiment, the operational data includes an inlet temperature and an outlet temperature of the one or more catalytic reactors (preferably each of the one or more catalytic reactors), an inlet pressure and an outlet pressure of the one or more catalytic reactors (preferably each of the one or more catalytic reactors), and a composition of the reaction mixture at the inlet and outlet of the one or more catalytic reactors (preferably each of the one or more catalytic reactors).

In one embodiment particularly suited for a styrene production plant, the operational data includes steam to oil ratio (STO), Liquid Hourly Space Velocity (LHSV), total styrene production normalized to catalyst volume, target ethylbenzene conversion, styrene selectivity, average inlet temperature, normalized pressure after the last catalytic reactor, normalized pressure drop across one or more catalytic reactors, temperature loss across one or more catalytic reactors, normalized deviation of temperature loss from expected values (calculated based on the conversion). The advantage of such a set of parameters is that it is still interpretable, since many parameters correspond to actual operating parameters or can be easily interpreted in this context. Other methods that reduce the number of problem dimensions (e.g., PCA or RFA) may produce parameters useful for modeling process and prediction accuracy, but often lack interpretability.

The data-driven model preferably performs a time series forecast and can cover the complete deactivation process of the catalyst during the entire production run in the reactor system. The model enables plant operators to refine and optimize operating strategies on a daily basis based on energy costs, market supply of raw materials or demand for plant products, and other constraints that may arise for the plant (e.g., interruptions to different plant sections or utilities).

In one embodiment, the forecast period spans the remaining life of the catalyst as determined by the limitations of the operating conditions (preferably the maximum reactor temperature that can be operated). The model allows the remaining life of the catalyst in use to be determined based on forecasted operating conditions and limits of those operating conditions (e.g., maximum reactor temperature or pressure at which operation may be performed). This enables the operator to reliably plan the time remaining until the next catalyst change, and also to simulate different operating strategies in order to extend catalyst life (if needed), or to achieve maximum productivity from the time remaining until the next planned plant shutdown.

In one embodiment, the data-driven model is used to predict operating conditions of catalysts in a production plant that do not provide historical data for training the data-driven model.

In one embodiment, the output of the data-driven model is used to optimize the operating conditions of a predetermined or current production run of the chemical production plant. In such embodiments, target operating parameters determined for more than one operating condition via the data-driven model are received and fed into the optimization processing device. The optimization may include one or more optimization objectives. The optimization goal may be specified by, for example, optimization parameters or goal parameters to be optimized. In this context, the optimization goal may further include finding a minimum or maximum value of the specified optimization parameter or target parameter. In addition to target parameters for more than one operating condition, optimization targets may be received and fed to the optimization processing device. For example, the operator may be provided with a selection of possible optimization objectives and the selected optimization objectives may be received based on the user selection.

In the case of one or more optimization objectives, an optimal solution may exist and may be provided as a result of the optimization. Such an optimal solution may be provided to a system for monitoring and/or control. The optimal solution may be displayed on a display device or used to control a chemical production process. In the case of more than one optimization objective or multiobjective optimization, one or more optimal solutions may exist and may be provided as a result of the optimization. A plurality of optimal solutions may be provided to the system for monitoring and/or control and displayed on a display device. In this case, the operator of the plant may choose between a plurality of optimal solutions, which simplifies the decision making process in the complex case of operating a chemical production plant.

In another embodiment, the optimization processing device determines a minimum or maximum value of the target operating parameter or a minimum or maximum value of at least one optimization parameter derived from the target operating parameter based on the target operating parameter as output by the data-driven model. In one exemplary scenario, the remaining life of the catalyst may be an optimization parameter derived from the target operating parameters and finding the maximum value is the goal of the optimization process. Furthermore, constraints may be included in the optimization problem. Limitations on target operating parameters may include maximum reactor temperature, minimum selectivity, or minimum production per day. In another exemplary scenario, the throughput over the remaining life of the catalyst or until a predetermined point in time for catalyst replacement may be an optimization parameter derived from the target operating parameters and finding the maximum value is the target of the optimization process. Constraints on target operating parameters may include maximum reactor temperature, minimum selectivity, or minimum and optionally maximum production per day.

In another exemplary scenario, multiple targets may be combined. For example, during the life of the catalyst, the remaining life of the catalyst at production may be an optimization parameter derived from the target operating parameters, and finding the maximum remaining life combined with the maximum production is the goal of the optimization process. Constraints on target operating parameters may include maximum reactor temperature, minimum selectivity, or minimum production per day. In another example, the plurality of objectives further includes optimal timing of catalyst replacement further considering cost aspects such as remaining production, energy requirements, catalyst replacement expense, or a combination thereof. For multi-objective optimization, known pareto (pareto) optimization techniques may be used.

In one embodiment, a data-driven model is used to simulate expected catalyst performance, productivity (e.g., total amount of styrene produced over a period of time), energy demand (associated with, for example, reactor or steam reactant heating), and cost or profit (e.g., related to reactant costs, market price of product, energy costs), for selecting operational data or sets of operational parameters in a catalytic reactor. Here, yield, energy consumption, CO₂Emissions, costs, byproducts, catalyst replacement intervals, remaining life, or a combination thereof may be the target operating parameters or optimized parameters derived from the target operating parameters. Different constraints may be defined in the optimization, such as constraints on the operational data on which the determination of the target parameters is based.

In one embodiment, the data-driven model is used to simulate expected catalyst performance, productivity, and energy requirements for selected operational data sets that are not practically achievable in a given plant setting (e.g., lower pressure level (deeper vacuum), lower STO ratio, or additional reactors). This may help plant managers to better assess the economics of potential upgrades to the plant.

In one embodiment, the data-driven model is used to simulate and/or evaluate the expected performance and operating conditions of a new plant that has not previously used a catalyst.

The present disclosure also provides a computer program product configured to perform the methods of the present disclosure. In an embodiment, the computer program product is a computer program implemented in a chemical production plant or a styrene production plant, in particular in a computing unit (computer) integrated therein and/or connected thereto. In one embodiment, the computer program product is integrated into a dashboard of a chemical production plant or a styrene production plant.

In one embodiment, a computer program product includes an interface for inputting operating conditions of a chemical production plant. The operating conditions may be historical data of previous production runs or actual operating conditions of the chemical production plant currently in operation.

In one embodiment of the computer program product, input parameters or operating data for forecasting are manually uploaded into the computer program product and/or read by the computer program product from the process control system. In one embodiment, the input parameters or operational data are provided in a formatted data table. In one embodiment, a graphical user interface is provided for manually uploading data into a computer program product. In another embodiment, an application programming interface for uploading data into a computer program product is provided.

In one embodiment, a user of the computer program product provides all of the information specified above (e.g., raw data, corresponding units, plant metadata) to the computer program product. This information may be entered manually, uploaded via a structured data file, or provided via an application programming interface (manual or automatic). For long-term prediction, the input time series of target operating parameters preferably covers at least the range required by the time-lag structure used in the model. If their time lags are also used in the model, control parameters may also be included.

In one embodiment, the exogenous operating parameters are input by a user, such as for a plurality of operating scenarios for which performance is to be predicted. All additional data processing (including formatting, aggregation, and prediction) is performed by the software product.

In one embodiment, the forecasts generated by the computer program product are presented to the user via a graphical user interface as a structured data file or via an application programming interface.

According to further embodiments, the computer program product is a computer program product which, when loaded into a memory of a computing device and executed by at least one processor of the computing device, performs the steps of the computer-implemented method described above.

The computer program product may be used with or incorporated into a computer system, which may be a stand-alone unit or include one or more remote terminals or devices that communicate with a central computer via a network, such as, for example, the internet or an intranet. Thus, the computers or processors and related components described herein may be part of a local computer system or a remote computer or an online system or a combination thereof. Any of the databases and computer program products described herein may be stored in a computer internal memory or a non-transitory computer readable medium.

Another aspect of the present disclosure is a computer system for predicting the performance of a catalyst in a chemical production plant or for determining the operating conditions of a chemical production plant. The computer system includes: at least one interface component configured to access and read operating parameters or operating data and catalyst specific parameters, in particular catalyst age indicators; and a processor unit implementing the data driven model and configured to predict performance of the catalyst by providing reactor operating parameters or operating data to the data driven model and catalyst specific parameters provided via the interface component. In one embodiment, a computer system is configured to couple to a chemical production plant including a catalytic reactor via a wired and/or wireless communication connection and to automatically access and read reactor operating parameters or operating data and/or catalyst specific parameters at least in part from a process control system of the chemical production plant including the catalytic reactor via an interface component.

Another aspect of the present disclosure is a computer-implemented method for training a machine learning-based data-driven model for predicting or forecasting performance of a catalyst in a catalytic reactor, including one or more catalytic reactor chemical production plants. The method includes providing a mathematical model as an initial basis; providing historical data, such as multiple production runs from the same type of catalyst and/or multiple production runs from multiple chemical production plants or catalytic reactors comprising the same type of catalyst, and optionally predetermined target operating parameters or operating and performance parameters of the chemical production plants (and in particular the catalytic reactors); accessing and importing the provided historical data into a mathematical model; adapting the parameterization of the data-driven mathematical model to the provided historical data; providing an update to the data-driven mathematical model on the basis of the adapted parameterization; and iteratively repeating the method steps by setting the updated data-driven mathematical model to an initial basis.

Another aspect of the present disclosure is a computer-implemented method for determining an operating and/or performance parameter or a target operating parameter of a chemical production plant including one or more catalytic reactors. The method includes accessing sensor data indicative of operating conditions present in the reactor system and catalyst-specific sensor data indicative of a catalyst currently used in the reactor system, determining operating and/or performance parameters or target operating parameters of the chemical production plant using a data-driven model, wherein the data-driven model is parameterized according to a training data set, wherein the training data set includes historical data, for example, from multiple production runs of the same type of catalyst and/or from multiple production runs in multiple catalytic reactors including the same type of catalyst and previously determined operating and performance parameters or target operating parameters, and providing the determined operating and/or performance parameters or target operating parameters of the chemical production plant including one or more catalytic reactors.

Another aspect of the present disclosure is a computer-implemented method for determining operational and/or performance parameters of a chemical production plant including one or more catalytic reactors. The method includes accessing sensor data indicative of operating conditions present in the reactor system and catalyst-specific sensor data indicative of a catalyst currently used in the reactor system, determining an operating and/or performance parameter or target operating parameter of the chemical production plant using a data-driven model, wherein the data-driven model is parameterized according to a training data set, wherein the training data set includes, for example, historical data from a plurality of production runs of the same type of catalyst and/or from a plurality of reactor systems including the same type of catalyst and a previously determined operating and performance parameter or target operating parameter of the chemical production plant, and providing the determined operating and/or performance parameter or target operating parameter of the chemical production plant.

Another aspect of the present disclosure is a control system for controlling a chemical production plant that includes one or more catalytic reactors. The control system comprises a computer system as described above and a control unit configured to control the actual and/or predetermined production run in the chemical production plant based on the provided operational and/or performance parameters or target operational parameters of the chemical production plant.

A computer program may be stored and/or distributed on a suitable medium, such as an optical storage medium or a solid-state medium supplied together with or as part of other hardware, but may also be distributed in other forms, such as via the internet or other wired or wireless telecommunication systems.

However, the computer program may also be presented via a network like the world wide web and may be downloaded into the working memory of a data processor from such a network.

According to another exemplary embodiment of the present invention, a data carrier or data storage medium for making a computer program available for download is provided, the computer program being arranged to perform the method according to one of the aforementioned embodiments of the present invention.

It is to be understood that the embodiments described herein are not mutually exclusive of one another, and that one or more of the described embodiments may be combined in various ways, as will be appreciated by one of ordinary skill in the art.

A computer program for performing any of the methods of the present invention can be stored on a computer readable storage medium (e.g., a non-transitory computer readable storage medium). The computer readable storage medium may be a floppy disk, a hard disk, a CD (compact disc), a DVD (digital versatile disc), a USB (universal serial bus) drive, a RAM (random access memory), a ROM (read only memory), and an EPROM (erasable programmable read only memory). The computer readable medium may also be a data communication network allowing downloading of the program code, e.g. the internet. The methods, systems, and devices described herein may be implemented as software in a digital signal processor DSP, microcontroller, or any other side processor, or as hardware circuits within an application specific integrated circuit ASIC, CPLD, FPGA, or other suitable device. The present invention may be implemented in digital electronic circuitry, or in computer hardware, firmware, software, or in combinations of them, e.g., in available hardware of a conventional mobile device, or in new hardware dedicated to processing the methods described herein, as will be described in more detail below.

Drawings

Exemplary embodiments of the invention are illustrated in the drawings. It is to be noted, however, that the appended drawings illustrate only specific embodiments of this invention and are therefore not to be considered limiting of its scope. The invention may include other equally effective embodiments.

FIG. 1 illustrates an exemplary embodiment of a method for determining changing operating conditions of a scheduled or current production run of a chemical production plant including at least one catalytic reactor;

FIG. 2 illustrates a high level workflow of pre-processing raw data from a single plant into a data set ready for model training or prediction;

FIG. 3 illustrates an example workflow for identifying raw data set selections underlying a model;

FIG. 4 illustrates an exemplary implementation of a system for determining changing operating conditions of a scheduled production run or a current production run of a chemical production plant.

Detailed Description

The present disclosure provides a computer-implemented method of predicting short-term performance of a catalyst and/or predicting long-term performance of a catalyst (including catalyst aging effects) in a chemical production plant including at least one catalytic reactor. The method involves the use of a mathematical model, in particular of a catalytic reactor, which is based on machine learning, does not involve prior information of the reaction kinetics and uses input parameters selected from the group consisting of sensor raw data, derived parameters, reactor operating parameters, plant metadata and parameters indicative of catalyst aging. The methods, systems, computer programs, and computer program products disclosed herein are further described with respect to a styrene production plant as an example. The methods, systems, computer programs, and computer program products disclosed herein are applicable to other production plants having at least one catalytic reactor, and in particular, having a fixed bed reactor.

In the case of a styrene production plant, a feed stream comprising ethylbenzene is mixed with steam in a mixer. The mixed stream is fed to a catalytic reactor that includes a potassium promoted iron oxide based catalyst for dehydrogenation to styrene monomer. The styrene production plant further includes temperature sensors, pressure sensors, flow sensors, etc. located at various locations to monitor plant operation.

During the production of styrene monomer, ethylbenzene may be dehydrogenated in an adiabatic radial flow reactor. Ethylbenzene is mixed with steam in a specific ratio called the boil-off gas (STO) ratio to provide heat for the endothermic dehydrogenation process and to prevent the potassium used in the process from promoting the reduction and coking of the iron oxide based catalyst. The reaction is run at elevated temperature and sub-atmospheric pressure in an arrangement typically comprising at least two successive reactors. Intermediate reheating may compensate for the energy consumed by the reaction. The low pressure, steam dilution and high temperature favor the dehydrogenation of ethylbenzene, resulting in higher equilibrium conversion.

During the life of the catalyst, potassium, which is a coke gasification promoter, evaporates from the catalyst and is carried downstream of the catalyst bed to its cooler outlet, resulting in a loss of activity caused by coking of the catalyst. To compensate for catalyst aging, plant operators increased the inlet temperature over the life of the catalyst to maintain constant ethylbenzene conversion. Higher temperatures negatively affect the selectivity to styrene by enhancing cracking and the formation of by-products (e.g., benzene and toluene). In addition, potassium deposits within the catalyst bed and fines generated by the catalyst result in an increase in reactor inlet pressure, which is thermodynamically unfavorable. When the inlet temperature or pressure has increased beyond the operating limits of the plant, the operation must be stopped and the catalyst needs to be replaced.

The catalyst performance and its aging rate depend on the reactor operating parameters such as inlet temperature, STO ratio, inlet and outlet pressures, ethylbenzene flow, the number of reactors used, etc. The process economics can be significantly improved by optimization of the reactor operation. Attempts have been made in the past by using different types of reactor models that have been fitted to catalyst operating data derived from experimental or industrial reactors. The model is based on knowledge or assumptions about the reaction kinetics of the primary and secondary reactions, mass, heat and pulse transport phenomena, adsorption/desorption of the different chemicals present in the system, coke gasification and potassium loss kinetics, etc. The methods, systems, computer programs and computer program products described herein allow for more robust and reliable process control.

FIG. 1 illustrates an exemplary embodiment of a method for determining changing operating conditions of a scheduled or current production run of a chemical production plant including at least one catalytic reactor.

If the determination relates to a predetermined production run, operation data indicating predefined operation conditions for the predetermined production run is received via the communication interface in a first step 10. The predefined operational data may be generated from a previous production run. If the determination relates to a change of the current production run, measured operation data indicative of current operation conditions for the current production run is received via the communication interface in a first step 10. Further, at least one operating data point may be adjusted such that it includes a desired operating value that indicates a change in current operating conditions. The operational data may include sensor data measured by sensors installed in the chemical production plant, quantities derived directly or indirectly from such sensor data, analytical data measured in samples taken from the chemical production plant, and/or quantities derived directly or indirectly from such analytical data.

Further, a catalyst age indicator associated with a period of time that the catalyst has been in use in a current or scheduled production run is received via the communication interface. The catalyst age indicator may be based on a point in time, a period of time, an amount derived from time-dependent operating data, and/or an amount cumulatively derived from time-dependent operating data. Further, plant metadata indicative of a physical plant layout may be received via the communication interface.

In a second step 12, the operational data and the plant metadata may be preprocessed via the processing device before determining the at least one target operational parameter. Preferably, the pre-processing includes conversion to a quantity that is independent of the physical plant layout.

In a third step 14, at least one target operating parameter for a predetermined production run or a changing operating condition of a current production run is determined via the processing device based on the operating data and the catalyst age indicator using the data-driven model. The determination of the at least one target operating parameter may additionally be based on plant metadata. The data-driven model is parameterized according to a training data set. The training data set may be based on a historical data set including operating data, catalyst aging indicators, at least one target operating parameter, and optionally plant metadata. The historical data set may include data from multiple runs, multiple plants, and/or multiple catalyst batches. The determination of the at least one target operating parameter may be based on a short-term model determining the target operating parameter at discrete points in time or a long-term model determining the target operating parameter over a period of time.

In the case of a long-term model, a time series of at least one target operating parameter measured, predicted or derived during a current or previous production run up to a predicted start point in time may be received in step 10 via the communication interface. Determining (14), via the processing unit, at least one target operating parameter at one or more time points subsequent to the predicted start time point may be based on the operating data including the desired operating value, a time series of the at least one target operating parameter, the catalyst age indicator, and optionally the plant metadata using a data-driven model. Preferably, the data-driven model includes an inherent time dependence.

In a fourth step 16, at least one target operating parameter for a predetermined production run or a changing operating condition of a current production run may be provided via the communication interface.

In a fifth step 18, the determined target operating conditions may be provided to an optimization process facility for optimizing the operating conditions of a predetermined production run or a change in a current production run of the chemical production plant. In this case, the determined target operating parameter can be received via the communication interface between the optimization processing device and the processing device for more than one operating condition of the predetermined production run or of a change of the current production run. Based on the received target operating parameters, a minimum or maximum value of the target operating parameters or a minimum or maximum value of optimization parameters derived from the target operating parameters may be determined via the optimization processing device, and a minimum or maximum value of optimal operating conditions indicative of a change in the predetermined or current production run may be provided via the communication interface.

Preprocessing of operational data

FIG. 2 illustrates an exemplary workflow of preprocessing data into a format suitable for model training, prediction via short-term models, or prediction via long-term models.

In a first step, measured operational data is optionally received, followed by a pre-processing method. If the data is ready for training, such operational data may include historical data sets from multiple runs, multiple plants, and/or multiple catalyst batches. Such operational data may include measured operational data indicative of current operating conditions of the current production run if the data is ready for prediction or forecasting. The operational data preferably comprises sensor data measured by sensors installed in the styrene production plant and/or analytical data measured in samples taken from the styrene production plant.

For each production plant, there are many sensors available, often hundreds or even thousands, that provide raw data at their respective sampling rates. Furthermore, analytical data (e.g., results of gas chromatography) may be used at specific times of sampling from the plant. The frequency at which this data is available varies from plant to plant, but is typically once a day, up to once a week.

In a first pre-processing step, the operating parameters may be selected from the operating data forming the input parameters of the data driven model. These input parameters may be derived from raw parameters, such as sensor data or analytical data. An exemplary process of how these original parameters are selected from all available parameters when setting up the data drive model is outlined in more detail below.

In a second preprocessing step, the data from the analyzed and selected sensors may be combined based on their time stamps and in particular preprocessed onto a common time scale. For many plants, only a daily summary of raw sensor data may be available, rather than high frequency raw data, and this is also the typical frequency with which analytical data is available, and a daily average may be used to acquire all data on a time basis. Other merging techniques may also be applied and are well known, such as interpolation of daily data (both analytics and sensors) and sampling of higher or lower frequencies of interest, for example to create hourly data.

Further, plant metadata may be received. Plant metadata (e.g., catalyst active volume and number of reactors) may be added to the data set as numerical or categorical variables to complete a set of input parameters for the model, including derived parameters.

In a third preprocessing step, the received and selected operational data may be filtered and smoothed. Here, for example, the time points and durations of maintenance intervals, start-up phases, irregularities and outliers can be identified and optionally filtered. To achieve this, many options are known to the person skilled in the art, and any combination of such methods may be used. For example, the procedure includes applying feasible absolute thresholds based on knowledge in the catalyst art, such as minimum reactor temperature, maximum steam/oil ratio, and maximum pressure after the reactor; the use of an absolute threshold, for example based on a distribution indicator, such as 6 times the interquartile range (conservative threshold used in the example application), or alternatively based on an estimated likelihood of data points originating from the overall distribution, identifies outliers by comparing each value or set of values to a distribution of all other values available from the same parameter or set of parameters of the corresponding production plant; and/or identifying irregularities based on a parameter jumping substantially compared to a monthly coefficient of change for the parameter.

In a fourth preprocessing step, missing data points on a common time scale may be detected and replaced with statistically determined values. Such a possibly missing parameter can be estimated. In particular, if these data have been sampled at a frequency lower than the selected time base, the analysis data may be interpolated. The interpolation may be determined by different methods, such as simple mean interpolation, forward or backward filling, weighted mean or estimated values from a kalman filter, or comparable estimation methods. The same method can be applied instead of the abnormal value. The start of the run can be identified based on expert defined criteria, for example as the first data point for a space velocity per hour >0.2/h in the styrene catalyst example. All derived parameters may then be calculated, including, for example, cumulative plant production.

Depending on the nature of the derived parameters, in particular the cumulative parameters, the identified start-up phases, downtime, etc. may be removed from the data set before or after the respective derived parameters are calculated. In an exemplary embodiment, the cumulative production since the start of the run (upon which the realized aging indicators are based) is calculated before the startup phase is removed from the dataset, as they contribute to the aging of the catalyst, even though these phases are not part of the operating conditions covered by the model.

Another pre-processing step not depicted in fig. 2 may be to convert data from different plants into common units. This operation is preferably and directly performed prior to initiating the workflow in fig. 2, although it may also be performed before or after any step in the process.

At this point, the data is ready for training, forecasting with long-term models, or prediction with short-term models. However, since filtering may cause some gaps in the data, the timescale for forecasting is typically much longer (e.g., months) than the available time base (e.g., days), and additional aggregation steps (e.g., weekly mean or median aggregation) may be performed on the data before it is used in the long-term model.

Parameter selection process

The operational data or raw parameters of interest may be determined via the workflow outlined in fig. 3. The parameters available to the catalyst expert serve as a basis, consisting of a combination of raw sensor data, analytical data and some typical operating parameters derived from them (e.g. conversion, space velocity, steam/oil ratio, selectivity). In a first step, if data from multiple plants is used, only the parameters available to all the plants of interest can be selected (if only one plant is modeled, the criteria are outdated), and the redundant parameters (e.g., readjusted parameters) and the zero variance parameters may be deleted.

Next, a correlation matrix for the remaining parameters may be computed, and clustering (e.g., hierarchical clustering) of the parameters may be performed to identify pairs or clusters of parameters that carry similar information. Instead of clustering algorithms, simple filters for high (anti-) correlation values (e.g., (>0.90 or < -0.90) or (>0.95 or < -0.95)) may be applied. However, clustering of the correlation value with all other parameters further allows identification of parameters that may have a lower direct correlation but very similar correlation values with all other parameters. From each cluster of two or more parameters, only a single parameter may be retained based on some selection criteria, such as: this parameter needs to be available to all plants; it is necessary to retain the typical parameters that the operator uses on a daily basis (this is an exception that may lead to the retention of more than a single parameter) and this should represent an interpretable quantity, which is why the dimensionality is not reduced via principal component transformations or similar methods that produce features that the operator cannot directly interpret.

Based on these criteria, the parameter set can be iteratively reduced by adjusting the clustering threshold as necessary. The number of parameters may be iteratively reduced until a set of parameters (1, …, M) is retained, wherein only low correlation between parameters is maintained. Some of the remaining relatively high correlation values stem from parameters that may be important to the operator and therefore may not be removed.

Once the parameter set is reduced, the raw sensor and analytical data are identified, which is necessary to obtain all of these remaining parameters.

Normalization of parameters and plant metadata

Different plants vary in their production levels, their typical operating conditions, and often exhibit systematic differences. To account for this difference between different plants, one or a combination of two strategies may be applied as described below.

In a further pre-processing step, normalization may be performed. In one embodiment, plant metadata may be received that indicates a physical plant layout. Such plant metadata may include reactor layout, such as number of reactors, active catalyst volume, reactor type, size, or combinations thereof.

Normalization of the operational data may be performed to make the model input parameters (other than plant metadata) as independent as possible from a particular plant layout. No normalization is required for many parameters such as steam/oil ratio, conversion, selectivity. In addition, parameters such as Liquid Hourly Space Velocity (LHSV) are essentially normalized by catalyst volume. Additionally or alternatively, the total or cumulative total yield of a catalyst production run per catalyst can be normalized by the active catalyst volume, as this is a more comparable measure of "aging" per unit volume of catalyst compared to the on-stream time (variation in production levels is not captured) or the non-normalized cumulative production (will have a different meaning for different sized reactors).

Additionally or alternatively, the pressure may be normalized to its initial value during the start of the run, e.g., the median of 90 days before commissioning in order to focus on aging effects, rather than (in some cases more significant) plant-to-plant differences. Additionally or alternatively, the pressure drop across each unit (e.g., reactor or heat exchanger) can be normalized by the space velocity or total flow rate, as it is well known that it varies with the superficial linear velocity of the gas mixture. Additionally or alternatively, an average inlet temperature of the reactor greater than one (e.g., 2 or 3) may be determined.

Finally, there are other reasonable ways in which operational data, analytical data, and any quantities derived therefrom may be compared between plants, in addition to the examples mentioned in this section; and possibly other useful reactor/plant metadata, particularly regarding reactor geometries that may be used in a similar manner.

The reason for finding such normalization parameters is that the use of data from multiple different plants provides a number of significant advantages: 1) the aggregate data set covers a parameter space much larger than that of any single plant, the latter often surrounding a relatively narrow set of operating parameters familiar to the operator, as erroneously deviating from this may result in significant monetary losses. Thus, a model trained on such an aggregate data set may provide predictions that are outside the operating range of a particular plant because it includes information that is not available from their own historical data. 2) Due to the long service life (2-3 years) of the exemplary styrene catalyst, each plant can only run 1-4 times a certain type of catalyst, severely limiting the number of deactivation processes that can be observed per plant (each run providing only a single independent observation of catalyst deactivation). Aggregating the data sets allows more deactivations to be included in the training data. 3) In combination with a parameter selection process that emphasizes the interpretability and availability of common operating parameters, this allows the application of trained models to plants that do not have a priori actual data available, a situation often encountered when technical recommendations need to be provided for new plants, which would not be possible if the model were strictly linked to a particular set of sensors available at a particular plant.

Short term model

For short-term prediction of catalyst behavior, any regression model may be used and various typical candidate models are known to those skilled in the art. Depending on the nature of the data in a chemical production plant, a small number of independent runs may typically be selected, and even after normalization of many parameters, some potential plant-specific deviation, relatively simple in nature and less flexible models may be selected. In such scenarios with a small number of independent runs, a highly flexible model (e.g., random forest regression) will fit the training data set better, but may not extrapolate well to the new data. Thus, depending on the number of runs available, an appropriate type or regression-based model combination may be selected.

One possible model may be a set of linear models trained on a subset of the training data, for example, to predict average reactor inlet temperature and selectivity of the reaction. The use of such a set has two advantages: first, using a set of models for prediction can result in more accurate predictions, e.g., by taking the median prediction of all models [ e.g., set methods — basis and algorithm, zhouximab; CRC publisher 2012 ]; second, training the set in this manner can estimate the uncertainty of the model by using the prediction horizon (or the 10% and 90% percentiles, or any other prediction quantile range … …).

When training data from multiple catalyst batches, multiple runs, and/or multiple plants is used, the training and testing data sets may be split by individual runs. Additionally, or alternatively, the training and validation data sets may be split by a single plant or catalyst batch. For example, a random set of approximately 75% of the training set may be selected, and the parameters may optionally be normalized (to zero mean and unit variance) during pre-processing. The remaining data can be used as a validation data set to test the trained model.

Long-term model

For time series forecasting, a number of mathematical models can be used, ranging from autoregressive models to recurrent neural networks. Model requirements for an exemplary problem at hand include: 1) applicable to multivariate time series, i.e. to predict a number of endogenous (uncontrolled) parameters with long-term trends to be predicted, such as pressure, inlet temperature or selectivity; 2) integration of exogenous (controlled) parameters, i.e. highly influencing parameters such as steam/oil ratio, LHSV, target conversion that are known or will be controlled externally and therefore do not require model prediction.

A preferred embodiment of the method relates to a mathematical model that allows regularization to avoid overfitting, which may easily occur when multiple time lags are included in the mathematical model. In one embodiment of generating the mathematical model used in the method of the present disclosure, a collection of mathematical models is implemented, the integration being based primarily on a combination of a penalized linear model and a penalized vector autoregressive model with exogenous Variables (VARX). For an overview of the penalized VAR (X) model, including different Structured Regularization methods, see, e.g., [ arXiv:1508.07497v1(Nicholson et al, 2018, VARX-L: Structured Regularization for Large Vector Automation with Exogenous Variables) ] and references therein.

Training/testing splits are performed on short-term models between runs. All candidate models can be trained to predict endogenous variables at time t based on the history of exogenous variables at time t and endogenous variables up to a maximum time lag m (t-m, …, t-1). Applying the model iteratively step by step and using the predictions of endogenous variables as input to the next step in each new step allows prediction of any number of steps in advance.

The training procedure may be performed at the factory level using leave-one-out cross-validation. Here for N plants in the dataset, N training datasets (consisting of data from all other plants) may be generated, their parameters may optionally be normalized to have a mean of 0 and a comparable range of values, and the trained model may be evaluated on the omitted plant (validation set). Finally, model hyper-parameters (e.g., regularization parameters) may be selected, which provide the most robust performance for the N validation sets, e.g., as measured by the mean root mean square error predicted one step ahead, and the model may be trained on the complete training set.

Since the mathematical model used in the method of the present disclosure is based on machine learning, it must be trained with historical data from at least one production run in a chemical production plant comprising at least one catalytic reactor before being used to predict short term performance or predict long term performance including aging of catalysts for such reactor systems. In one embodiment of the method of the present disclosure, historical data from multiple production runs of the same type of catalyst has been used to train the model. In a further embodiment of the method, historical data from production runs in multiple reactor systems including the same type of catalyst has been used to train the model. In both embodiments, the historical data may be provided from different manufacturing batches of the same type of catalyst. In one embodiment of the method, the operational data and catalyst aging indicators for possible multiple production runs have been normalized before being used to train the model, as described in the previous section.

It has been found that using operational data from more than one production run, whether from the same plant or from different plants, can improve the quality of the prediction and broaden the range of operational parameters covered by the predictive or forecasting model. Including data from multiple runs and plants further increases the popularity of forecasting or forecasting to apply to production plants where no data is available during model training. However, all data used to train the model is preferably provided by the plant using the same catalyst formulation, as different catalysts differ significantly in their catalytic properties (reaction rates) and morphological properties (transport properties). This includes using the same type of catalyst provided in different manufacturing lots or on different delivery dates.

In one embodiment, the input parameters for the model are selected from sensor data available from the reactor system, such as analytical data from Gas Chromatography (GC) analysis, and derived parameters listed in the corresponding section above.

In one embodiment, the operational data or reactor operating parameters include the inlet and outlet temperatures of each reactor, the inlet and outlet pressures of each reactor, and the composition of the reaction mixture at the inlet and outlet of each reactor.

In one embodiment, the operational data or input parameters for the model include the boil-off gas (STO) ratio, Liquid Hourly Space Velocity (LHSV), total styrene production normalized to catalyst volume, target ethylbenzene conversion, styrene selectivity, average inlet temperature, normalized pressure after the last reactor, normalized pressure drop for the reactor, temperature loss for the reactor, normalized deviation of temperature loss from expected (calculated based on conversion) values.

An advantage of the parameter set used in the method or system of the present disclosure is that it is still interpretable, as many parameters correspond to actual operating parameters or can be easily interpreted in this context. Other methods of reducing the problem dimension (e.g., PCA or RFA) may produce parameters useful for the modeling process and prediction accuracy, but often lack interpretability.

The mathematical model used in the method of the present disclosure performs time series forecasting and can encompass the complete deactivation process of the catalyst throughout the production run in the reactor system. The model enables plant operators to refine and optimize operating strategies on a daily basis based on energy costs, market supply of raw materials or demand for plant products, and other constraints that may occur in the plant, such as interruptions in different plant sections or utilities.

The subject matter of the present disclosure is further described and explained in the following working examples.

Examples of the invention

Generation of data sets for training mathematical models

To develop the mathematical model, data from multiple production runs of the same catalyst type (BASF S6-42) was used. This data set covers information for 11 industrial plants with 2 or 3 reactors and about 1-4 production runs per plant. For each individual plant, an entire set of parameters (including analytical data and sensor data) is collected. Sensor data is typically provided at daily resolution, while analytical data is provided at daily to weekly resolution.

The parameter selection process of the model is as described above (FIG. 3)

Table 1 lists the operational data from the raw sensors and the analytical parameters that were selected at each time point in order to derive all relevant parameters of the model. These unit and format choices are just one example that may be used; the temperature may also be specified, for example, in degrees fahrenheit, the pressure may be specified in mmHg, and another date and time format may be used, etc.

TABLE 1

The number of reactors and the total catalyst volume are additionally used as metadata. Table 2 below lists the parameter sets (derived parameters and plant metadata) used to train the different models.

TABLE 2

An exemplary application: short term model

Model development

As described above, a set of 50 linear regression models has been trained on subsets of the training data set, each subset being split between runs. Each of the 50 reduced training sets contains approximately 74% of the available runs randomly selected to improve the prediction and provide an estimate of the local uncertainty about the model prediction (assuming each model is trained on a different subset of operating conditions, for example). Importantly, the data subsets need to be split between runs (or alternatively between plants) rather than a random sampling of the training data points. Otherwise, there will hardly be any variation between the models, since all models are trained on almost the same distribution of operating conditions.

In the present example, all other parameters in table 2 have been used in particular to predict the parameters "temperature" (average reactor inlet temperature) and "selectivity" (reaction to the desired product styrene).

In this particular implementation, the reaction temperature itself is one of the main influencing factors of the selectivity observed. Thus, the prediction of the two parameters is performed in two steps. First, a set of 50 models was trained to predict temperature based on all parameters in table 2 (excluding selectivity and temperature). Using all the parameters in table 2 as inputs (excluding selectivity only), a second set of 50 models was trained to predict selectivity. The second set is then used to predict selectivity using the predicted temperatures from the first set of models as one of the input parameters.

In practical use of the developed model, the workflow of subsequently predicting both parameters has been implemented in a single prediction function that receives the input parameters and predicts both temperature and selectivity. Thus, for all uses of the model, the combination of the two sets of models, viewed externally, can be considered as a single entity of the "short-term" model.

In addition to the direct use of the parameters in table 2, the interaction terms are also taken into account, and finally, for example, the "steam/oil ratio-conversion" interaction is integrated into a short-term model of the predicted temperature. Such interactions or higher order polynomial terms (e.g., quadratic terms for selectively predicted temperatures) can be easily implemented into the predictive model using a statistical programming language without extending the basic set of parameters provided to the model.

The choice of which type of model, whether or not to use aggregation techniques, and which higher-order terms, transformations, or interactions of input parameters to use, depends on the particular problem and dataset; and is a typical model development program for data scientists.

Short term model case 1

In one exemplary use case, a plant operator may wish to use a short-term model to estimate a required reactor temperature to achieve a desired target conversion at specific operating conditions that may not have been used before their plant. In the case of a styrene production plant, such operating conditions may be increased steam-to-oil ratio or decreased LHSV or changes in feed composition. Ideally, the starting point for this prediction should be the current plant state, including the normalized aging parameter "totalProduction".

Model implementation

FIG. 4 illustrates a client server arrangement of a production monitoring and/or control system, including a client having a processing device containing a user application, a server having a processing device for a service provider. The client and server sides may be communicatively coupled, e.g., wired or wirelessly, via a communication interface. The client may include a display device. Preferably, the client user application is configured to receive and display operational data, desired values, catalyst aging indicators, plant metadata, target operating parameters, or determined operating conditions. More preferably, the client user application is configured to receive the target operating parameter or the determined operating condition to control a current or predetermined production run in the chemical production plant based on the target operating parameter or the determined operating condition. The client user application may be an embedded part of a chemical production plant process monitoring and/or control system.

Since the start of a production run, raw sensor and analytical data are recorded at the production plant along with the necessary plant metadata. These data are preprocessed according to a data preprocessing workflow, for example as shown in fig. 2, in particular using the same workflow (filtering step, interpolation, thresholding, aggregation, etc.) as used for preparing the training data set for the respective model. Preferably, this preprocessing is implemented by the same party that has developed the model and may be provided to the user application, for example, via direct integration of preprocessing functions or via an Application Programming Interface (API).

After the data has been pre-processed, the user may adjust the operating parameter of interest to provide a desired operating value that indicates a change in the current operating conditions. The set of adjusted input parameters (also referred to as an operating scenario or a prediction scenario) is then transmitted to the prediction function. The prediction function may be implemented locally, for example in a user application, or addressable via an API, and perform all of the operations described above. The results will be reported back to the user application, for example, for comparison and selection between different scenarios. The user may further associate the input parameters and/or predicted parameters with, for example, costs or other quantities that may affect the decision process regarding which operating parameters to use in the plant.

FIG. 4 visualizes an exemplary implementation concept. The raw data is collected automatically or manually at the production plant and transmitted to the service provider via an API that processes the raw data into the correct format according to a workflow, for example, as described above. The transformed data set may be provided to a user (e.g., a plant operator) and different scenarios may be defined based on the current values. These scenarios may be transmitted to the same or a second API of the operational model, which provides corresponding predictions or forecasts to the user.

Short term model case 2

In a second use case, an expert may wish to provide an estimate of the reactor temperature and corresponding selectivity development prior to plant installation of the catalyst. This scenario often occurs during preparation of a technical proposal that provides one or more hypothetical operational scenarios and their impact on customers, for example, before making a decision to purchase a catalyst. Therefore, accurate prediction of a plant that never provided data for model training is desired.

Some simplifying assumptions are made, such as estimating the "aging" parameter totalProduction based on assumed operational data, typically involving constant operating parameters for a complete operation, replacing the temperature loss of the reactor with estimates based on operating conditions, etc., and the above-described short-term model may be used for this purpose in view of a typical set of operating parameters provided by the customer. Such services that may be provided by the model are a direct result of the selection criteria used in the parameter selection workflow (fig. 3), particularly keeping all parameters as inputs to the model, which operators typically use to monitor and control their plants.

Model presentation to a user

The interface of the user application may contain an input parameter block of the model, which may be automatically or manually populated, and may be used to interactively specify an operational scenario; a model output block for a given scenario (e.g., average reactor inlet temperature and selectivity with a predicted range); and optionally as further output of a model to predict a local response to a parameter of interest (e.g., a predicted reactor inlet temperature for a range of target conversions).

The input parameter block may be automatically populated if current plant operational data is available (e.g., use case 1) or may be fully manually defined if no actual plant data is available (e.g., use case 2). Based on the user-adjusted input parameters, the parameters predicted by the model, in this example the reactor inlet temperature and styrene selectivity, can be displayed to the user, for example, as text (for predicting a single point), or for example, in graphical form (response of the local model to a single parameter change). These are used only as examples, as there are more ways in which the model can be used using use cases, implementations, and presentation to model users.

An exemplary application: long-term model

Model development

Starting with the process data set (Table 2) used to develop the short-term model, some additional steps are taken to prepare the data set for training the long-term model. First, a smaller number of parameters are selected from the list, and second, each parameter is aggregated weekly.

The input parameters of the model may be constant (transactions, Catvolume), may be controlled (conversion, SOR, LHSV), may be calculated based on these constant or controlled values (totalProduction), or may be uncontrolled in the operational scenario of interest (temperature, selectivity, pressureOut, deltaP, deltaT, dTdev _ norm).

The latter set of parameters are labeled as endogenous or uncontrolled parameters, while the former (known throughout the prediction, assuming the operational scenario is performed as planned) are labeled as exogenous or controlled parameters.

Table 3 provides the parameters used to develop an exemplary forecasting model and its assignment to both types of parameters.

Parameter name	Exogenous/endogenous sources	Type (B)
			reactors	External source	Is absolute
temperature	Endogenous source of	Numerical value
			pressureOut	Endogenous source of	Numerical value
deltaT	Endogenous source of	Numerical value
			SOR	External source	Numerical value
LHSV	External source	Numerical value
			conversion	External source	Numerical value
selectivity	Endogenous source of	Numerical value
			CatVolume	External source	Numerical value
deltaP	Endogenous source of	Numerical value
			totalProduction	External source	Numerical value
dTdev_norm	Endogenous source of	Numerical value

TABLE 4

In developing the exemplary model, many different VARX-type candidate models as described above are trained, where the candidate models differ, for example, in their regularization methods (elastic net, ridge) or their maximum number of time lags (4-10 weeks).

For all of these different model combinations, the training procedure was performed in the same manner at the plant level using leave-one-out cross-validation. All trained models are candidate models that ultimately comprise a set of long-term models.

Although all training is performed for one-step forecasting ahead (which can be applied iteratively, predicting any number of steps), the model ultimately selected should also perform well for longer forecasts. To evaluate this, predictions are initially performed at multiple points of each available run in the training and testing dataset using the actual exogenous variables of these runs as operational scenarios, and the error distribution of the forecasts is determined in advance for each step.

For reactor inlet temperature and selectivity on both training and test data sets, a model was chosen that did not provide significant long-term bias and narrow error distributions over the entire range (even though modest broadening of the longer predicted error distributions had to be expected). Model selection is based on the error distribution of predictions performed on the test and training set from 30 candidate models.

In this exemplary development of the long-term model of styrene catalyst, the model finally selected was 3 two-phase models with a maximum time lag of 10 weeks trained with different types of regularization on the local ramp. For ensemble prediction, all models are run iteratively independently of each other, and only the complete individual predictions are averaged to provide an ensemble prediction. In the presented example, this is a more efficient implementation than aggregating forecasts after each individual step, which would be a possible alternative implementation of aggregate prediction.

As with the single model, the final set is evaluated on the test data to obtain an error distribution for the next N days, an exemplary implementation provided as a forecast error estimate.

As with the short-term example above, ensemble forecasting and expected error distributions together may be considered by the user as a single entity of a "long-term" model for model applications, regardless of the details of the underlying program (ensemble-averaged aging, multi-step model....).

Long term model use case

In an exemplary use case, a plant operator may wish to estimate the remaining catalyst life depending on different scenarios of operating the plant and plant-specific limits on operating parameters. Such scenarios may include changes in LHSV, steam/oil ratio, or target conversion levels; there are many motivations to consider different scenarios, but an exemplary issue would be whether it would ultimately be worthwhile to run the plant at lower production levels during the low styrene price phase to extend catalyst life.

The end of catalyst life may depend on a number of conditions defined locally at each production plant, but one limitation of catalyst life is reactor temperature or pressure which in all cases may not exceed a plant-specific threshold. From the predictions of the long-term model, the end of catalyst life may be estimated based on the threshold and compared for different user-defined scenarios.

Model implementation

The model may be implemented conceptually similar to that described in FIG. 4, with some minor variations. Processing functions, implemented locally or via an API, for example, need to be adapted to provide a data format for long-term model training, in this particular example a weekly aggregation step after executing a program on data for a short-term model. The prediction function, implemented locally or via an API for example, now receives not only a set of operating conditions to predict a single point, but also the lag endogenous data required by the model, as well as the operating scenario values for the exogenous parameters. Similarly, the model output is a complete forecast of all or only part of the endogenous variables of the model.

Model rendering for a user

In an implementation of the long-term model, the endogenous parameters used in the model need to be available at least for the last L weeks, where L is the maximum time lag used in any model element included in the "long-term" model. This data may be obtained automatically from the plant raw data as described in the data preprocessing and preparation section, or may be manually entered or uploaded in an appropriate format.

The operator may use the control parameter input block to create an extended period of operating scenarios, for example, in an example application, the steam/oil ratio, target conversion, and LHSV may be planned ahead of time in up to three separate sections to also simulate future changes. These scenarios may also be much more complex than those described herein. The planned operational scenario may be visually displayed, preferably along with a history of these parameters to more easily maintain some control continuity. The control parameters selected for user manipulation may consist of any subset of all exogenous parameters used in the model. In this particular implementation, the other 3 exogenous parameters are constant, or may be derived directly from other inputs and/or times.

The forecasts provided by the long-term model may be displayed in tabular format, ready for export of data, and based on further analysis of the forecasted trends, for example, or the results may be graphically displayed to the user for visual inspection of one or more different planning scenarios.

The interface may comprise a section in which selected control (exogenous) parameters may be defined for user-defined operating scenarios (e.g., target conversion, LHSV and steam/oil ratio; graphs representing actual data up to the start of the prediction and operating scenarios for the future). In another portion of the interface, the user may be presented with selected endogenous parameters (e.g., reactor inlet temperature and styrene selectivity; the start of prediction may be marked with a horizontal dashed line, beyond which everything is predicted by the model and includes an estimate of the prediction error). Furthermore, predictions of all or some of the endogenous parameters may be presented as data boxes that may be derived by the user.