System Research On Dynamic Trend Prediction Of Batch Process Based On LSTM And Attention Mechanism

With the advancement of industrial modernization,higher requirements are put forward for the safe and stable operation of batch production process.As an important industrial production mode,batch process has the characteristics of unsteady,nonlinear and time-varying.Therefore,how to accurately predict the dynamic operation trend of batch production process,find out the abnormal situation as soon as possible,and take corresponding treatment measures,so as to ensure the normal operation of industrial production process,has become a research hotspot and technical problem to be solved in this field.To solve this problem,the theoretical research and application development of the system are carried out based on the deep learning time series model,The main work is as follows:Firstly,aiming at the low accuracy of LSTM(Long-Short Term Memory)network in dynamic trend prediction of batch production process,a time series prediction model datlg based on the fusion of TG-LSTM(Transformation-Gated LSTM),GRU(Gated Recurrent Unit)and two-stage attention is proposed.The encoder adopts TG-LSTM network,which can enhance the representation of short-term dependence while maintaining long-term dependence,and accurately capture local mutation information in time series;The decoder uses GRU network to realize the fast decoding of information while maintaining the accuracy;At the same time,the attention mechanism is introduced to realize the comprehensive and accurate description of time series feature information in the two dimensions of space and time.The data set of penicillin fermentation process was selected to train and test the datlg prediction model,and the good performance of the prediction model was verified.Secondly,because the dynamic trend of penicillin fermentation batch process is complex and changeable,and the variables are closely related,it is difficult to refine and decode the trend of the complex process only by using single-layer GRU network.A time series prediction model based on TG-LSTM-double-layer GRU and two-stage attention is proposed.The double-layer GRU network structure is used to replace the single-layer GRU network to fully interpret the coding information,so as to further improve the prediction accuracy of the model.Experiments on penicillin fermentation process data set also verify the good prediction performance and anti-interference ability of the prediction model.Finally,the dynamic trend prediction system of batch process based on LSTM variants and attention fusion is developed.According to the above theoretical research and prediction model,the application system of time series prediction based on deep learning is analyzed,designed and implemented.The main functions include the introduction of application scenarios,theoretical interpretation of prediction model and time series prediction function,which lays a good foundation for the practical application of dynamic trend prediction of batch process.Through the above theoretical and applied research,the dynamic trend prediction model of batch process based on the combination of LSTM variant and two-stage attention and the developed application system further improve and expand the modeling and prediction method of complex batch process,and provide new solutions and application system options for ensuring the safe and stable operation of industrial production process.