TE process time sequence prediction method based on transfer entropy and long short-term memory network

Abstract

The invention relates to a TE process time sequence prediction method based on transfer entropy and a long short-term memory network. In order to solve the problems of low time sequence prediction precision and low training speed caused by high relevance among TE process variables and easiness in introducing redundant information into a prediction model, the asymmetry of a transfer entropy algorithm is used for variable selection, an upstream variable which greatly affects the temperature of a reactor is selected from TE process reactor unit variables, and the interference of downstream irrelevant variables is eliminated, so that the complexity of a time sequence prediction model is reduced. By using the excellent performance of the LSTM in the aspect of time sequence prediction, an LSTM time sequence prediction model is established based on a variable selected by the transfer entropy, and the future time sequence of the temperature of the reactor is predicted.

Description

technical field [0001] The present invention relates to the technical field of key variable selection based on information entropy and time series prediction based on deep learning, in particular, a method based on transfer entropy and long short-term memory network is proposed for the characteristics of strong correlation between TE (Tennessee Eastman) process variables. The TE process timing prediction method is an important branch in the field of process industry technology. Background technique [0002] The purpose of sequence prediction of industrial processes is that on-site operators can control the entire production process by monitoring key process variables, thereby ensuring the safety and smooth operation of the production process. Performance and quality indicators of optimized products and safe operation of industrial processes play an important role. [0003] LSTM (Long Short-Term Memory), which is changed from RNN, introduces various gate structures, which gr...

AI Technical Summary

Problems solved by technology

[0004] Aiming at the problems that the TE process variables are strongly correlated and redundant information is easily introduced into the prediction model, resulting in low timing prediction accuracy and slow training rate, a TE process timing prediction method based on transfer entropy and long short-term memory network is proposed.

Images