Research On Risk Traceability And Prediction Method Of Complex Chemical Process Based On Deep Learning

Modern chemical process presents the development trend of large-scale and complex devices,and the following dangerous and hidden dangers of chemical process are also gradually increasing.It is difficult to accurately describe its operation status and fault conditions by traditional methods through complex and complicated process parameters of chemical process.In view of the above problems,this paper combines complex network theory with deep learning,and makes use of the advantages of complex network theory in studying the structure and complexity of the system and the advantages of deep learning in processing and analyzing the nonlinear characteristics of large-scale data to carry out risk traceability and prediction research on chemical processes.The research is mainly carried out from the following four aspects:(1)Risk characterization modeling of chemical process.A modeling method of chemical process risk representation based on Spearman-Apriori is proposed.Based on multi-source process data,Spearman correlation analysis and Apriori association rule mining are carried out on variable data to build a complex network model of chemical process(undirected right directed right);According to the characteristic parameters of complex network and the law of network node degree distribution under different working conditions,the chemical process risk is characterized and the risk source is determined.(2)Deep traceability of chemical process risks.A tracing method of chemical process risk evolution based on community structure is proposed,which is based on the directed weighted complex network model of chemical process;After Deep Sparse Autoencoder sparse processing,Louvain is used to divide the community structure,sort according to the node importance,and then trace the risk evolution path of the chemical process in combination with the process flow.(3)Chemical process risk tracking and prediction.The risk prediction and analysis of chemical process based on network structure entropy is proposed,which is based on the directed weighted complex network model of chemical process;Divide the data in the form of window and calculate the network structure entropy,normalize its Max-Min dispersion to obtain the relative risk value sequence,and determine the chemical process system risk according to the distribution law of the relative risk value under different working conditions;Through the training and prediction of Attention-Bi LSTM model,the risk of chemical process is tracked and predicted.(4)Chemical process risk deduction platform.Based on PyQt5,a comprehensive risk management and control platform for chemical process is designed and developed.The platform layout is designed in combination with Qt Designer and the data visualization window is designed by using Pychart.The platform integrates the risk deduction process of modeling,traceability and prediction,and takes TE process as an example to monitor the risk of chemical process through data visualization.