基于Deep-LSTM的工业过程监测方法应用研究

AbAbstractThe modern process industry process system with streamline production process,productioncondition,production equipments,production control automation,closely related to all parts of thecomponent such as the industrial characteristics,and process industry process with nonlinear regression,the characteristics of sampling data made the process industry system anomaly detection and faultdiagnosis problem become more prominent,and in the event of failure,will cause serious property lossand casualties,seriously hinder the process of industrialization and economic development.In order toreduce the risk of process industrial failure and effectively avoid the loss of personnel and property,process industrial process fault prediction technology came into being.Currently,the predictionmethods based on time series data include traditional statistical analysis method,deep learningrepresented by cyclic neural network (RNN),machine learning and other prediction methods.It is difficult to accurately predict the faults of complex industrial processes by the abovemethods.Combined with modern process industry sampling data of the nonlinear process principle andprocess industries,the features of sequence data is not stable,based on the depth of the length of thememory network time series prediction monitoring model to realize fault prediction effectively,bychanging the length of the intemal structure of memory network,in a moment LSTM Cell can havemultiple LSTM units,enhance the long-term dependence and feature extraction ability of the networkstructure,on the basis of the single-layer network structure at the same time,increase the network layer,in ensuring maximum sampling data extracted on the basis of the characteristics of temporal and spatialcharacteristics,And enhances the network structure of dependency for a long time,can more effectivelydeal with complex process,large amount of data of time series of industrial time series problem,according to the characters of the extracted data to establish the temporal relationship between data,combined with univariate statistical control method,set up control limit,to real-time monitoring ofpredictor variable values,according to predict whether a variable's value transfinite judge whether thereis a failure,if the predicted value transfinite,namely failure immediately generate alarm information,accomplish fault prediction and alarm in advance,can effectively avoid the happening of processindustry process fault,to ensure the safety of the process of process industry production.In TE (Tennessee Eastman)test on the chemical process simulation platform,multiple depth isstudied by using TE simulation data set,neural network model of a typical fault disturbance analysiscontrast experiment,the experimental results show that the time series of Deep-LSTM failureprediction process industry process monitoring model can effectively predict failure process industryprocess,at the same time also has higher precision of predi