Research On Model Temperature Prediction Based On Recurrent Neural Network

In the process of low-pressure casting,the data has the characteristics of time ductility,uncertainty and nonlinearity.The current research methods can not solve the problem of time delay,which makes it difficult to establish an accurate temperature prediction model of low-pressure casting production mold.The accurate prediction of mold temperature in low-pressure casting production process provides the basis for the automatic control of low-pressure casting.Therefore,the prediction of mold temperature in low-pressure casting production process is of great significance.Aiming at the problem of time-delay in low-pressure casting process,a multivariable time long and short term memory network(MT-LSTM)and a multivariable time gated recurrent unit network(MT-GRU)model were proposed,and fireworks algorithm(FWA)was used The model parameters of MT-LSTM and MT-GRU were optimized by algorithm.The accurate prediction of mold temperature and the automatic optimization of model parameters were realized.The specific research work is as follows:Firstly,by studying the process of low-pressure casting,the candidate variables that affect the mold temperature are analyzed.The information acquisition system platform was built to collect the mold temperature data and the candidate influencing variables which affect the mold temperature and store them in the database.The gray correlation analysis method was used to analyze the correlation degree between the variables,and the correlation analysis between the input variables and the mold temperature was transformed into the gray correlation degree between the variables.Thus,the key variables affecting the mold temperature were obtained,which laid a foundation for the establishment of the prediction model of mold temperature in low-pressure casting production.Secondly,the sliding window method is used to map the time series data including time delay to the input layer of recurrent neural network(RNN),and MT-LSTM and MT-GRU are established.By learning the law of time delay,the problem that time delay cannot be determined is solved,and the influence of time delay on model temperature prediction is eliminated,and the prediction accuracy of the model is improved.Finally,to solve the problem that it is difficult to select the optimal parameters of deep learning model,FWA is used to optimize the parameters of MT-LSTM and MT-GRU respectively to get rid of the interference of artificial experience in the selection of model parameters.The experimental results show that the proposed method has high accuracy and strong generalization ability,and can realize the accurate prediction of mold temperature and automatic optimization of model parameters.