Research On Time-delay Pearson Correlation Analysis And Key Variable Prediction Method For Air Separation Equipment

Air separation system belongs to a variety of process industry.The manufacturing technique of air separation is to radically change air into liquid nation by means of repeated pressurization and freezing,etc.The ensuing liquid air is distilled to separate the desired gases.In the production process of air separation system,to ensure smooth operation of the production process,the field staff will adjust the relevant equipment according to the changes of certain indicators.However,the actual production process is not able to detect the production of some key variables in time,which may cause some hidden problems to the whole production process.How to effectively and accurately predict the key variables in the air separation system has become an important issue.With the development of the process industry,air separation equipment,which has a complex mechanism,the mechanism-based modeling approach will require greater costs,and accuracy and reliability may not be guaranteed.Data-driven prediction methods only use a large amount of input data to construct a model that relates input and output variables,without focusing on mechanistic information and with accurate and efficient prediction results.Therefore,this paper analyzes the correlation between the variables of the air separation system from this perspective through a large amount of field production data and performs effective model predictions.The main work of this paper is as follows:(1)The traditional correlation analysis method between variables often does not take into account the time lag between variables when processing variable data,and air separation systems generally have massive time lags,so the effectiveness of the traditional correlation evaluation technique will be reduced.Therefore,this paper proposes a characteristic determination method based totally on the time-lagged Pearson correlation coefficient,which is enhanced by incorporating the time lag thing and Hurst index evaluation based totally on the Pearson correlation coefficient,and verifies the superiority of the improved approach by applying it to air separation equipment.(2)By introducing multiple neural networks for prediction models,the optimal correlated set is selected for key variables in air separation equipment using a hybrid variable selection-prediction method,and simulation experiments are conducted on the correlation variable sets obtained from multiple correlation analysis methods and multiple neural network models.The results show that the improved Pearson correlation analysis method combined with the Long Short Term Memory network(LSTM)model provides more accurate prediction results for key variables in air separation equipment.(3)In order to improve the accuracy of the prediction model,the key variables are also processed based on the improved empirical modal decomposition algorithm,and the features of the input time series data are abstractly extracted by the convolutional neural network.Finally,the model is predicted by the Gated Recurrent Units(GRU).The results show that the key variable signals after the improved empirical modal decomposition algorithm can have more significant modal components with minimal reconstruction errors,and the Convolutional Neural Networks(CNN)and GRU model provides prediction results that are closer to the real values.