Application Of Long-short Term Memory Network In Model Predictive Control

By analyzing the control problems related to pH neutralization process,this paper designs an identification method and control strategy based on long-term memory network and dynamic matrix control.This reaction process is involved in many industrial production,such as chemical production,wastewater treatment,life pharmacy and other fields.Therefore,it is very important to control pH quickly,accurately and stably significance.This kind of controlled object has several obvious characteristics:first of all,the reaction process of this kind of object is complex and has strong nonlinear characteristics;and this kind of object will have large time lag characteristics in practice;therefore,the identification accuracy achieved by traditional modeling method is low,which is not conducive to the implementation of later control strategy;in addition,there are many unknown disturbances,which will control the system The system brings some difficulties.For this kind of system,the traditional PID control strategy is used to achieve poor control effect,while the use of modern control theory will cause the control failure because of the poor model accuracy.Therefore,this paper adopts the predictive control method which not only can online control but also has low requirements for model accuracy.In this paper,the previous research results are integrated,the reaction process is carefully analyzed,and the predictive control theory is summarized.On this basis,by designing a new network topology structure,combining the advantages of short-term memory network and common neural network,the process model can be better established,and on the basis of the network identification model,the piecewise linear dynamic matrix control is designed The algorithm is designed,the relevant control scheme is designed,and the computer simulation experiment is carried out.This research mainly includes the following aspects:In this paper,a new network structure is designed,that is,the LSTM unit and the ordinary neuron are fused into a hidden layer.Through this special structure,the network can solve the nonlinear and time-delay problems at the same time.Through the comparative experiment,the results show that the method proposed in this paper has smaller identification error and faster learning speed than other methods Rate,better identification effect and better generalization ability.On the basis of the proposed identification model,we propose a piecewise linear dynamic matrix control strategy.The method first obtains the piecewise vector model of the system through the network model,and further obtains the dynamic matrix,designs the performance index,rolling optimization and feedback correction formula,and conducts effective control of the process and Simulation Research.Compared with other methods,the reliability and effectiveness of the algorithm are proved.The simulation results show that the proposed method can make the system achieve the control objectives quickly and accurately,and has good robustness.