Research On Compound Optimization Control Strategy Within Batch And Between Batches Of Batch Chemical Process Based On Data-driven Technique

This paper is devoted to the research of intelligent optimization control method for batch production process of special fine chemicals.The production control of special fine chemicals belongs to batch process control.The production process has strong nonlinear and batch repetition characteristics.At present,the classical PID control strategy is used in the production.PID control has the characteristics of good reliability and simple maintenance,but it is difficult to meet the requirements of high-precision control of complex process.The product quality and operation process of previous special chemicals production was unstable,indicating that PID control strategy is difficult to meet the control requirements of complex process.How to improve the existing PID control strategy and make full use of the repetitive characteristics of batch production in the case of difficult to establish an accurate mathematical model,how to directly use online and offline data to realize the optimization control of complex production process and equipment,which is the research focus of this paper.In this paper,a compound control method is adopted for batch production processes,such as Chylla Haase reaction process.The control of batch process is divided into two dimensions: within batch control and between batches control.The compound control strategy is designed,in which intelligent self-tuning PID control within batch is combined with ILC between batches.The repetitive characteristics between batches are fully utilized,and the adaptive improvement of control is realized within and between batches.For the control within batch,PID control architecture is adopted,and LM optimization algorithm is used to realize the self-tuning of PID control parameters.RBF neural network is used to identify the Jacobian information generated in the optimization process,and an improved DE algorithm is used to optimize the initial value of PID self-tuning parameters,the center and width of radial basis function and the initial connection value of neurons.The control strategy within batch does not require to obtain the mathematical model of the controlled object,only takes the process data as the control source,which has high practical value.For the control between batches,P-type iterative learning control which has practical application value is adopted to suppress the repeative disturbance.In order to realize the data-driven adaptive improvement of this control method,a strategy of unfalsified control with limited parameter set is designed.This control strategy not only realizes the function of suppressing repeative disturbance between batches,but also has practical adaptive adjustment ability,which is superior to the fixed parameter iterative learning control method.In the production process of special fine chemical D1,the core quality control part is distillation and purification process.Due to the complexity of the internal mechanism of batch distillation process,The type or composition of the pre separated mixture varies frequently.According to chaos theory,temperature can reflect the reaction and separation in the system.Accurately judging the fractionation point is the key to the production of special fine chemical D1.Based on the reality,data-driven modeling is implemented.A LSTM neural network prediction model is established to predict the fractionation point.The structure of LSTM neural network is complex,so it needs to optimize the parameters.Bayesian optimization algorithm is designed to optimize the parameters.The prediction model of LSTM neural network is established to realize the prediction of fractionation point.In D1 production process,purity data is regarded as the key index and only detected at the end of production.An iterative learning control algorithm for end-point quality based on BP neural network is designed.Firstly,BP neural network is used to establish the prediction model of production process variables and end-point product purity.Based on the neural network prediction model,the iterative learning control of end-point purity is implemented,the satisfactory results show that the quality control of batch distillation process with batch repetition is realized...