The Research On Correlation Analysis And Prediction Methods Of Production Process Variables In Process Industry

In the production process of process industry,a number of key variables or indexes are an important manifestation of stable and safe production.Field personnel need to monitor certain variables of special concern,so as to real-time control the entire production process,and finally achieve the goal of stable operation.However,for the real process industry,some important variables can not be monitored or comprehensively monitored in real time,which makes the real-time control of the whole production process difficult to achieve.Therefore,it is particularly important to establish an effective and stable prediction model of production process variables for real-time prediction of production process variables.Data-driven method only needs to establish the relationship model of input-output variables in industrial process system to achieve the purpose of prediction,without paying attention to the mechanism information of production process.Therefore,this paper conducted in-depth research on feature selection methods and modeling methods in data-driven production process variable prediction modeling.The specific research work is as follows:(1)The traditional feature selection methods based on correlation analysis does not consider the influence of time-delay when analyzing the correlation of variables,so that the accuracy of the analysis results is seriously reduced,which makes it impossible to select the input variables effectively,which is not conducive to the subsequent predictive modeling.Thus a grey relational analysis method based on cross-correlation analysis is proposed.The time-delay is determined by the time lag calculation method based on cross-correlation,and then this is used as an input parameter to the gray relational analysis method.The method calculates the correlation degree between variables and selects the variables most relevant to the target variable.Effectively eliminates the interference of time-delay to variable correlation analysis.The effectiveness of the improved method was verified by the application of specific case.(2)A threshold comparison strategy is defined for the problem of correlation variable selection and redundant variable deletion in feature selection.Combined with correlation-based feature selection method,the optimal subset of feature variables is selected.Applying it to practical cases in the process industry,the results show that the improved method is practical and effective in solving such problems.(3)Taking a specific process industrial production system as the research background,the improved grey relational analysis and mutual information analysis method are combined with Nonlinear Autoregressive with External Input(NARX)neural network,Long Short-Term Memory-Recurrent Neural Network(LSTM-RNN)to establish a predictive model of product purity.By comparing and analyzing the prediction performance of the four models,it is found that the prediction accuracy of the LSTM-RNN model combined with the improved gray relational analysis method,the stability of the prediction results,and the data tracking effect are better than other models.The applicability and validity of the improved grey relational analysis method was verified by the application of actual cases,and the performance advantages of LSTM-RNN model were highlighted.