Application Research Of Industrial Process Monitoring Method Based On Deep-LSTM

With the development trend of modern process industry,such as the expansion of production scale,the upsizing of production equipment,the automation of production control and the complexity of process flow,and the non-linear and autoregressive characteristics of the process sampling data,the problem of abnormal detection and fault diagnosis in process industry system becomes more difficult.However,once a certain link or device fails,it will spread to other relevant parts,causing serious property loss and casualties,and seriously hindering the process of industrialization and the development of social economy.For early detection process in industrial production process of abnormal or fault,timely alarm and disposal,and to eliminate the potential hazard,ensure the safety of the production running smoothly,try to avoid personnel and property losses,as a kind of effective solutions,industrial process monitoring process has become the industry research hot spot,has the important theory significance and application value.Aiming at the monitoring problems existing in the complex industrial production process,based on the time-varying and nonlinear characteristics of modem industrial production process and process sampling data,firstly,the LSTM model of long and short time memory network is used to model,predict and analyze the sampled data,so as to extract the inherent characteristics and variation trend of the sequence.Secondly,in view of the LSTM model can't characteristics of time series data information to long-term memory and the shortage of transmission,memory model,when depth of length,are in every LSTM Cell increase LSTM unit number to enhance the capacity of feature extraction and information memory,at the same time deepening of network layer,in ensuring maximum extraction sampling data,on the basis of temporal and spatial characteristics,further enhance the network structure of dependency for a long time,can effectively deal with industrial production time series.Then,by combining the Deep-Lstm time series prediction model with the single-variable statistical process monitoring Shewhart chart,a complete monitoring method flow is presented to realize the online monitoring of the production process.Thirdly,taking Tennessee Eastman chemical process as an example,RNN,GRU,LSTM and Deep-Lstm models were used to compare the prediction and alarm of step and random disturbance production processes.The results showed that the process monitoring method based on Deep-Lstm time series model had higher accuracy and stability.Finally,based on the monitoring method,a corresponding process industrial process monitoring system was designed and developed by using Python graphical interface development tool and combining with the application requirements of industrial process monitoring.The dynamic visualization of observed variables,process fault and their prediction,as well as the functions of fault warning,parameters and related Settings were realized.The above research and software development of Deep-Lstm time series prediction model provide a new solution for online process monitoring in the process industry--early fault detection and timely disposal.However,there are still some problems in the dynamic monitoring of the process of process industry,such as noise pollution and the extraction of the coupling relationship among multiple variables,which need to be further improved and perfected so as to better promote the healthy development of process industry.