阅读说明：本技术 一种基于模糊神经网络的氯乙烯精馏温度控制方法 (Chloroethylene rectification temperature control method based on fuzzy neural network ) 是由 于现军 吕伟军 陆晟波 于 2019-10-25 设计创作，主要内容包括：本发明公开了一种基于模糊神经网络的氯乙烯精馏温度控制方法,涉及氯乙烯精馏生产控制领域。首先通过信号采集模块对传感器信号进行离散采集,并对其进行滤波处理。其次,根据信号采集模块输出进行模糊神经网络控制,分别依次进行模糊化、模糊推理、模糊决策、解模糊化输出得到所需控制参数值,直接对执行器进行非线性解耦控制。最后通过BP学习算法,对控制器中的连接权值、高斯函数中心值和宽度值进行学习修正,使控制器输出后,精馏塔温度误差收敛,提高控制精度的基础上,提高系统的稳定性。(The invention discloses a chloroethylene rectification temperature control method based on a fuzzy neural network, and relates to the field of chloroethylene rectification production control. Firstly, discrete acquisition is carried out on the sensor signals through a signal acquisition module, and filtering processing is carried out on the sensor signals. Secondly, fuzzy neural network control is carried out according to the output of the signal acquisition module, fuzzification, fuzzy reasoning, fuzzy decision and defuzzification are respectively carried out in sequence to obtain required control parameter values, and nonlinear decoupling control is directly carried out on the actuator. And finally, learning and correcting the connection weight, the Gaussian function central value and the width value in the controller through a BP learning algorithm, so that the temperature error of the rectifying tower is converged after the controller outputs, and the stability of the system is improved on the basis of improving the control precision.)

1. A chloroethylene rectification temperature control method based on a fuzzy neural network comprises the steps of (1) a signal collector; (2) a fuzzy neural network controller; (3) learning an algorithm; the method is characterized in that: by adopting a method of combining fuzzy control and neural network control, on the basis of improving the control precision by carrying out nonlinear decoupling control on the neural network, the fuzzy control reduces the frequency and amplitude of regulation and improves the stability of the system.

2. The signal collector as claimed in claim 1, wherein in the control process, discrete sampling is performed on the feeding flow, the tower top temperature, the middle temperature, the tower bottom temperature and the tower top reflux amount through set sampling time, sliding average filtering is performed on the current sampling value and the past 5 continuous sampling values, the maximum value and the minimum value of 5 values of continuous sampling are removed, and the filtering value is obtained by summing and averaging the remaining 3 values.

3. The fuzzy neural network controller of claim 1, which uses a Mamdani model to obtain the required control parameter values by fuzzifying, fuzzy reasoning, fuzzy decision and defuzzification output of input signals, wherein a structural framework of the controller is built by adopting a multilayer neural network, connection weights of each layer are corrected by a learning algorithm, and the output of the controller directly controls an actuator.

4. The fuzzy layer respectively fuzzifies each parameter of the input layer, and each variable can be subjected to fuzzy segmentation of different levels according to the characteristics of the field device.

5. Each neuron node in the fuzzy inference layer represents a fuzzy rule in the traditional fuzzy control, and the output value of each neuron node corresponds to the matching degree of the current input and each fuzzy rule.

6. The fuzzy decision is parameter normalization calculation, and the number of the neuron nodes in the layer is the same as that of the fuzzy inference layer.

7. And the defuzzification output layer realizes accurate output of the controller model.

8. The learning algorithm of claim 1, wherein the connection weight, the gaussian function center value and the width value in the controller are learned and corrected by a BP error reverse transfer method, so that after the controller outputs, the temperature error of the rectifying tower is converged, and the stability of the system is improved on the basis of improving the control accuracy.

Technical Field

The invention relates to a vinyl chloride rectification temperature control method based on a fuzzy neural network, in particular to a temperature control method for a vinyl chloride rectification high-boiling tower.

Background

Vinyl Chloride Monomer (VCM) is one of the most important raw materials in the chemical industry, and the current VCM monomer used for producing polyvinyl chloride (PVC) accounts for more than 96% of the total world production. The purity of the vinyl chloride monomer has a direct relationship with the conversion rate of vinyl chloride produced by final PVC polymerization, and has an important influence on the quality of PVC polymerization products. The purification process of the vinyl chloride monomer after the generation reaction is mainly a rectification mode, a low-boiling tower heats a crude vinyl chloride raw material liquid to be below the boiling point of vinyl chloride, so that low-boiling impurities are separated from the raw material liquid, and then the raw material liquid is heated to the boiling point of vinyl chloride through a high-boiling tower and is condensed and separated from the top of the tower, so that the high-purity vinyl chloride monomer is obtained.

In the rectification process of chloroethylene, gas-liquid two phases exist in a tower and are mutually heat and mass transfer, the process is complex, the system is a typical multi-input multi-output system, has high input hysteresis, slow dynamic response and a high-order mathematical model, is strong in coupling, has serious nonlinear response, and is difficult to accurately measure and mathematically model the rectification process. Meanwhile, higher requirements are provided for the control of the rectifying tower, and fluctuation of tower conditions and instability of product quality are easily caused by fluctuation or unreasonable operation and adjustment of feed components. In the control indexes of the rectifying tower, the temperature index is particularly important, and the content of impurities in the final vinyl chloride monomer is directly influenced.

In the past, the control of the vinyl chloride rectifying tower is usually carried out by adopting single-loop PID control or decoupling control with feedforward, even some places still rely on the experience of operators to carry out manual operation, so that the control level still stays on the general normal operation of production maintenance, and the satisfactory purification effect is difficult to achieve. Therefore, how to eliminate the influence of system coupling and nonlinear response on control precision and stability is a problem to be solved in the control of the vinyl chloride rectification system.

Disclosure of Invention

In order to overcome the influence of multivariable coupling and nonlinear response of the existing vinyl chloride rectification system, the invention provides a vinyl chloride rectification temperature control method based on a fuzzy neural network.

In order to achieve the purpose, the invention adopts the following technical scheme: a chloroethylene rectification temperature control method based on a fuzzy neural network comprises a signal collector, a fuzzy neural network controller and a learning algorithm.

The signal collector is used for discretely sampling the feeding flow, the tower top temperature, the middle temperature, the tower kettle temperature and the tower top reflux amount through set sampling time in the control process, filtering, reducing the influence of interference signals on the controller, and outputting the interference signals to the controller.

The fuzzy neural network controller obtains required control parameter values by fuzzifying, fuzzy reasoning, fuzzy decision and defuzzification output of input signals by using a Mamdani model, a structural framework of the fuzzy neural network controller is built by adopting a multilayer neural network, connection weights of each layer are corrected by a learning algorithm, and the output of the fuzzy neural network controller directly controls an actuator.

The learning algorithm adjusts parameters in the fuzzy neural network controller by a BP error reverse transfer method, so that the system error is converged.

Due to the adoption of the technical scheme, the invention has the following advantages: 1. realizing the decoupling control of each variable of the system; 2. the control precision and the stability of the system are improved.

Drawings

FIG. 1 is a control flow diagram of the present invention;

fig. 2 is a diagram of a fuzzy neural network structure based on the Mandani model.

Detailed Description

The invention is further described below with reference to the accompanying drawings.

Referring to fig. 1, a vinyl chloride rectification temperature control method based on a fuzzy neural network comprises the following steps: (1) signal acquisition; (2) fuzzy neural network control; (3) and (5) learning an algorithm.

In the signal acquisition in the step (1), discrete sampling is carried out on the feeding flow, the tower top temperature, the middle temperature, the tower kettle temperature and the tower top reflux quantity at intervals of 10 seconds (one sampling period) by the signal acquisition device, the current sampling value and the past 5 continuous sampling values are subjected to sliding average filtering, the maximum value and the minimum value of the continuously sampled 5 values are removed, and the rest 3 values are subjected to summation average calculation to obtain a filtering value.

In the step (2), the fuzzy neural network has 5 layers in total, and as shown in fig. 2, the fuzzy neural network comprises an input layer, a fuzzy inference layer, a fuzzy decision layer and a defuzzification output layer.

The input layer feeds the material

And derivatives thereof

Temperature error at the top of the column

And derivatives thereofDeviation of the temperature at the top of the column from the temperature at the middle of the column

And derivatives thereof

Deviation of middle temperature from tower still temperature

And derivatives thereofTemperature of tower still

And derivatives thereof

Amount of reflux at the top of the column

And derivatives thereof

Are directly connected with each neuron node and are all

For discrete sampling values at a time, the input matrix is as follows:

. Wherein the content of the first and second substances,

，

the temperature at the top of the column is,

for the desired value of the top temperature,，

the temperature of the middle part is shown as the temperature of the middle part,

。

the fuzzy layer fuzzifies each parameter of the input layer respectively, each variable can be subjected to fuzzy segmentation of different levels according to the characteristics of field equipment and can be divided into positive large, positive middle, positive small, zero, negative small, negative middle and negative large, and corresponding membership function of the variable is calculated

. Wherein the content of the first and second substances,，

，

is composed of

The dimension(s) of (a) is,

is composed of

The number of fuzzy partitions of (1). Because the rectification system belongs to a slow time-varying process, in order to ensure the smooth and stable control, the following Gaussian function is used as the membership function:

wherein the content of the first and second substances,

and

respectively, the center value and the width value of the membership function.

Each neuron node in the fuzzy inference layer represents a fuzzy rule in the traditional fuzzy control, the output value of the neuron node corresponds to the matching degree of the current input and each fuzzy rule, and in order to facilitate mathematical calculation and expression, a formula is adopted:

. Wherein the content of the first and second substances,，

，

，，

。

the fuzzy decision is parameter normalization calculation, the number of neuron nodes in the layer is the same as that of a fuzzy inference layer, and the calculation formula is as follows:

，

。

the defuzzification output layer realizes accurate output of the controller model and has the expression

Wherein, in the step (A),

，，

，

is input by a reflux adjusting valve at the top of the tower,

the reboiler hot water regulating valve input.

The learning algorithm in the step (3) is to perform learning correction on the controller parameters, and when the fuzzy segmentation level of each input is determined, the parameters to be learned in the controller are the connection weight between the fuzzy decision layer and the defuzzification output layer

，

And in membership functions in the fuzzification layerHeart value

And width value

，

Solving the parameters by using a BP error gradient descent method to obtain:

wherein

Wherein