阅读说明：本技术 一种常压塔顶油干点在线分析检测方法 (Online analysis and detection method for oil dry point of atmospheric tower top ) 是由 陈夕松 迟慧 王鹤莹 沈煜佳 陈伟睿 梅彬 于 2020-05-18 设计创作，主要内容包括：本发明公开了一种常压塔顶油干点在线分析检测方法,该方法首先判断在线分析仪能否正常工作,若无法工作则根据工艺机理选择相关变量,然后通过软测量模型M<Sub>1</Sub>计算塔顶油干点测量值y<Sub>1</Sub>；再对待测变量和相关变量的历史数据进行主元分析,得到估计模型M<Sub>2</Sub>,计算塔顶油干点估计值y<Sub>2</Sub>；最后通过卡尔曼滤波融合y<Sub>1</Sub>和y<Sub>2</Sub>,得到塔顶油干点的最优估计值y。本发明提出了基于主元分析的塔顶油干点估计算法,并且将估计值与基于软测量模型的测量值相结合,能有效提高在线成分分析检测的准确性,对原油蒸馏过程中先进控制的实施发挥了重要作用。(The invention discloses an online analysis and detection method for oil dry points at the top of a normal pressure tower 1 Calculating the measured value y of the oil dry point at the top of the tower 1 (ii) a And performing principal component analysis on the historical data of the variable to be measured and the related variable to obtain an estimation model M 2 Calculating the estimated value y of the oil dry point at the tower top 2 (ii) a Finally fusing y through Kalman filtering 1 And y 2 And obtaining the optimal estimated value y of the oil dry point at the tower top. The invention provides a tower top oil dry point estimation algorithm based on principal component analysis, and combines an estimated value with a measured value based on a soft measurement model, so that the accuracy of on-line component analysis and detection can be effectively improved, and the method plays an important role in implementing advanced control in the crude oil distillation process.)

1. An online analysis and detection method for oil dry points at the top of an atmospheric tower is characterized in that an estimation model is established based on principal component analysis, and an estimated value of the oil dry points at the top of the atmospheric tower is fused with a measured value obtained through a soft measurement model to obtain an optimal estimated value, so that the online detection of the oil dry points at the top of the atmospheric tower is realized, and the online analysis and detection method comprises the following steps:

(1) judging whether the normal oil dry point online analyzer works normally, if so, directly measuring the dry point value, and turning to the step (9), and if not, turning to the step (2);

(2) selecting variables related to the variable y to be measured to form a vector x;

(3) obtaining a state space model M of the system through soft measurement calculation₁Calculating the measured value y of the variable y to be measured₁；

(4) Taking normal historical data of the variable y to be measured and the related variable as a training set to form a sample matrix X;

(5) performing principal component analysis on the sample matrix X to obtain a load matrix P;

(6) let the direction vector ξ of the variable to be measured_i＝[1 0 … 0]^TAnd calculates ξ_iProjection to residual subspace ξ ℃_i：

(7) Computing estimation model M₂：

y₂＝Cx (2)

Wherein, y₂As an estimate of the variable to be measured, I₁Is a matrix of the units,after being unitizedI₂Is a ratio of I₁A one-dimensional less identity matrix;

(8) fusing measurement value y by Kalman filtering₁And estimateEvaluating y₂Obtaining the optimal estimated value y of the variable;

(9) and (6) ending.

2. The method according to claim 1, wherein the variables related to the dry point of the atmospheric tower top oil are selected from the group consisting of atmospheric tower top temperature, atmospheric tower top pressure, atmospheric tower top cold reflux temperature, atmospheric tower middle layer gas phase temperature, atmospheric tower feeding temperature, top circulation reflux heat extraction ratio, and medium reflux heat extraction ratio, which are 7 variables.

3. The on-line analysis and detection method for the dry point of the atmospheric tower top oil according to claim 2, characterized in that the heat extraction ratio P of the top circulation reflux is P₁Can be circulated from the top of the atmospheric tower₁Constant-top circulating heat exchange temperature difference delta T of constant-pressure tower₁And calculating the feeding amount F to obtain:

4. the on-line analysis and detection method for the dry point of the atmospheric tower top oil according to claim 2, characterized in that the heat extraction ratio P of the first intermediate reflux is P₂The flow F can be returned from the normal₂Constant-pressure tower heat exchange temperature difference delta T₂And calculating the feeding amount F to obtain:

5. the on-line analysis and detection method for the oil dry point of the atmospheric tower top according to claim 1, characterized in that the model M is₁Can be obtained by combining a mechanism analysis method and least square regression.

Technical Field

The invention relates to industrial process state online detection, in particular to an online analysis and detection method for an oil dry point at the top of an atmospheric tower in a crude oil distillation process.

Background

In the advanced control of the crude oil distillation process, the multivariable control using quality indexes as controlled parameters is often used, so that the purposes of highest product yield, minimum processing energy consumption of a device and product quality edge clamping guarantee are achieved, and the economic benefit is maximized. Wherein, the common topping oil dry point is the main quality control index of the common topping product.

The common oil dry point is difficult to be directly measured by an instrument, most of the common oil dry points depend on off-line analysis and test, the real-time performance is poor, and the quality closed-loop control cannot be realized. An on-line analysis technology based on near infrared spectrum is developed later, the common top oil is sampled and heat exchanged to 40 ℃, and then spectrum data are analyzed so as to calculate dry points, but the near infrared spectrometer is high in cost and sometimes fails or is in a maintenance state. The dry point calculation based on soft measurement techniques may be based on a mechanistic model or a statistical model. The mechanism model is based on scientific principles such as physics and chemistry, and utilizes technologies such as material balance and energy balance to obtain a mathematical model of a common-overhead oil dry point and related process variables, but the establishment of the mechanism model requires deep knowledge of the crude oil distillation process and is accompanied with certain assumed conditions, so that the model precision is not high enough. The statistical model is obtained by collecting a great deal of process historical data and modeling by using a statistical analysis data processing method, but the actual calculation precision of the established model is reduced because the crude oil property and the operation condition are frequently changed.

Disclosure of Invention

Aiming at the problems, the invention discloses an online analysis and detection method for oil dry points at the top of an atmospheric tower, which is characterized in that an estimation model is established based on principal component analysis by selecting relevant variables, and the estimated value of the oil dry points at the top of the atmospheric tower is fused with the measured value obtained by a soft measurement model to obtain the optimal estimated value, so that the online detection of the oil dry points at the top of the atmospheric tower is realized.

The method comprises the following steps:

(1) judging whether the normal oil dry point online analyzer works normally, if so, directly measuring the dry point value, and turning to the step (9), and if not, turning to the step (2);

(2) selecting variables related to the variable to be detected to form a vector x;

(3) obtaining a state space model M of the system through soft measurement calculation₁Calculating the measured value y of the variable to be measured₁；

(4) Taking normal historical data of the variable to be measured and the related variable as a training set to form a sample matrix X;

(5) performing principal component analysis on the sample matrix X to obtain a load matrix P;

(6) let the direction vector ξ of the variable to be measured_i＝[1 0 … 0]^TAnd calculates ξ_iProjection into residual subspace

(7) Computing estimation model M₂：

y₂＝Cx (2)

Wherein, y₂As an estimate of the variable to be measured, I₁Is a matrix of the units,after being unitizedI₂Is a ratio of I₁A one-dimensional less identity matrix;

(8) using a Kalman filterFiltering the fused measured value y₁And the estimated value y₂Obtaining the optimal estimated value y of the variable;

(9) and (6) ending.

In the present method, selecting variables related to the overhead oil dry point includes: the top temperature of the atmospheric tower, the top pressure of the atmospheric tower, the cold reflux temperature at the top of the atmospheric tower, the gas phase temperature of the middle layer of the atmospheric tower, the feeding temperature of the atmospheric tower, the top circulating reflux heat extraction ratio and the first medium reflux heat extraction ratio are 7 variables in total.

In the method, the heat extraction ratio P of top circulation reflux₁Can be circulated from the top of the atmospheric tower₁Constant-top circulating heat exchange temperature difference delta T of constant-pressure tower₁And calculating the feeding amount F to obtain:

in the method, the heat extraction ratio P is obtained by refluxing₂The flow F can be returned from the normal₂Constant-pressure tower heat exchange temperature difference delta T₂And calculating the feeding amount F to obtain:

in the method, model M₁Can be obtained by combining a mechanism analysis method and least square regression.

Has the advantages that:

the online analysis and detection method for the atmospheric tower top oil dry point disclosed by the invention can estimate the tower top oil dry point by using an estimation model based on principal component analysis through the correlation among variables according to other measured correlation variables when an online analyzer cannot work, and comprehensively calculates the estimated value with a measured value obtained by soft measurement calculation to obtain an optimal estimated value, so that the result is more accurate, the advanced control engineering of the industrial process is effectively implemented, and the economic benefit of an enterprise is improved.

Drawings

FIG. 1 is a flow chart of an implementation of an online analysis and detection method for oil dry point at the top of an atmospheric tower

FIG. 2 is a schematic diagram of a process of atmospheric tower in a refinery

FIG. 3 is a comparison graph of online analysis and detection results of oil dry point at the top of atmospheric tower of a certain oil refining enterprise

Detailed description of the preferred embodiment

The following detailed computing process and specific operation flow are given in conjunction with the accompanying drawings and specific examples to further explain the present invention. The present embodiment is implemented on the premise of the technical solution of the present invention, but the scope of the present invention is not limited to the following embodiment.

In the case, a certain oil refining enterprise is taken as an example, in order to ensure the stable operation of the production process and reduce the cost while ensuring the product quality, the enterprise designs a plurality of advanced controllers for the crude oil atmospheric and vacuum distillation device, wherein the advanced controllers comprise a furnace efficiency controller, an atmospheric furnace branch balance controller, an atmospheric tower controller and the like. The constant-top oil dry point is a main quality control index of a constant-top product and is an important controlled variable in an atmospheric tower controller.

The effectiveness and the implementation process of the method are illustrated by the online analysis and detection of the oil dry point at the top of the atmospheric tower in the atmospheric and vacuum device of the enterprise. 500 sets of data were selected, with the first 400 as the training set and the last 100 as the test set to validate the method. The implementation flow of this case is shown in fig. 1, and the specific implementation steps are as follows:

firstly, from the process mechanism, the factors related to the common oil dry point are qualitatively analyzed, and the process schematic diagram of the atmospheric tower of the enterprise is shown in fig. 2. According to the specific process analysis, the dry point of the atmospheric top oil is related to 7 variables of the atmospheric tower top temperature, the atmospheric tower top pressure, the atmospheric tower top cold reflux temperature, the atmospheric tower middle layer gas phase temperature (in the case of 35 layers of the atmospheric tower), the atmospheric tower feeding temperature, the medium reflux heat extraction ratio and the top circulation reflux heat extraction ratio, as shown in table 1:

TABLE 1 Dry Point related variables for the Top product of atmospheric tower

Serial number	DCS position number	Description of variables
			1	TI1103A.PV	Top temperature of atmospheric tower
2	PI1112.PV	Top pressure of atmospheric tower
			3	TI1135.PV	Cold reflux temperature at top of atmospheric tower
4	TI1129.PV	35-layer gas phase temperature of atmospheric tower
			5	TI1139.PV	Atmospheric tower feed temperature
6	T2LOOP1.PV	Top circulation reflux heat extraction ratio
			7	T2LOOP2.PV	Heat extraction ratio of one medium reflux

One of the medium reflux heat extraction ratio and the top circulation reflux heat extraction ratio can be obtained by calculating the heat exchange temperature difference, the flow and the feeding amount, and the formula is shown in table 2:

TABLE 2 formula for intermediate variable calculation

Establishment of common-roof oil dry point soft measurement model M by combining mechanism analysis with partial least square method₁Written as a spatial state system model in the form:

x(t)＝Ax(t-1) (1)

y₁(t)＝Bx(t) (2)

wherein x (t) is the state vector at the time t, the vector is 7 × 1 and consists of the 7 related variables, and a state transition matrixy₁(t) is the common-ceiling oil dry point measurement at time t; state observation matrix B ═ 1.01290.51340.5967 … 0.9265]。

All 100 groups of data in the test set are substituted into a soft measurement model M₁The measurement result of the constant oil dry point is calculated and obtained as shown in FIG. 3a, and the mean square error MSE₁3.5360, the maximum error is 6.0544.

Then, 400 groups of samples in the training set form a training matrix X, and principal component analysis is performed to obtain 6 number of principal components and 8 × 6P dimension of the load matrix, as shown in formula (3):

forming 100 groups of test set data into a test matrix, and enabling the direction vector ξ of the variable to be tested_i＝[1 0 0 … 0]^TSubstituting the load matrix P into the calculation ξ_iProjection into residual subspaceAs shown in formula (4):

calculation model M₂As shown in formulas (5) and (6):

y₂＝Cx (5)

according to model M₂Calculating to obtain the variable value y to be measured₂Mean square error MSE, as shown in FIG. 3b₂2.6731, the maximum error is 4.8317.

Will y₁(t) and y₂(t) performing iterative computation according to Kalman filtering formulas shown in formulas (7) to (12):

x(t|t-1)＝Ax(t-1) (7)

P(t|t-1)＝AP(t-1)A^T(8)

K(t)＝P(t|t-1)C^T[CP(t|t-1)C^T]^-1(9)

x(t)＝x(t|t-1)+K(t)(y(t)-Cx(t|t-1)) (10)

y(t)＝Cx(t) (11)

P(t)＝(I-K(t)C)P(t|t-1) (12)

wherein the matrices A, C are respectively model M₁、M₂Is calculated in the above step, let the initial value y (0) be 170, P (0) be a 7 × 7 dimensional matrix with all elements 0.1, x (t | t-1) be a state vector x (t-1) based on the time t-1, according to the model M₁Recursion is carried out to obtain a t-moment state vector value; p (t | t-1) is a state vector covariance matrix corresponding to x (t | t-1); p (t) is the corresponding covariance matrix of x (t); k (t) is a Kalman gain matrix at time t; x (t) is the optimal estimation of the state vector at time tCounting; and y (t) is the optimal estimation of the constant top oil dry point at the time t.

The optimal estimation value of the common-top oil dry point is finally obtained after Kalman filtering, as shown in FIG. 3c, the mean square error MSE₃1.0526, the maximum error is 3.3645.

The error comparison of the above three results is shown in Table 3, the mean square error and maximum error ratio model M of the method₁And M₂The result is small, so the method can more accurately obtain the variable value to be measured and meet the process control requirement.

TABLE 3 error comparison

Model (model)	Mean square error	Maximum error
			M1	3.5360	6.0544
M2	2.6731	4.8317
			Fusion value	1.0526	3.3645