Research On Data Driven Modeling And Prediction Methods Of Air Separation Process

The air separation process is a kind of process industry.The production process is continuous and the mechanism is complicated,some of the key indicator variable data are important manifestations of the equipment and process conditions.The industrial field personnel usually adjust the operation of related equipment according to the change trend of key indicators to ensure that the production process is in a stable running state.However,in the actual production environment,some important indicators are difficult to detect in real time or comprehensively,which will negatively affect the overall working condition adjustment and planning and scheduling.Therefore,it is essential to establish a model that can realize the short-term prediction of the production process indicator variable.The control systems of the modern industrial production site can collect a large amount of data containing production information,this article mainly focuses on the correlation analysis,forecasting model and multi-step forecasting method in the data-driven based air separation process variable forecasting modeling.The specific research work is listed as follows:(1)The traditional correlation analysis methods use real-time data corresponding to the variables to analyze the correlation degree between variables.When there is a time delay between the variables,the accuracy of the calculation results will be seriously reduced.Therefore,this paper proposed a grey correlation analysis method based on the time-delay calculation of the maximum information coefficient.The timedelay between variables is obtained by the maximum information coefficient method and integrated into the grey correlation model as a parameter to determine the degrees of grey incidence between variables.Furthermore,the accuracy of the improved correlation analysis method was verified through the application of the air separation production process that affected by the time delay factors.(2)The indicator variables of the production process are related to multiple process variables.In order to obtain the optimal set of related variables of the air separation system indicator variable,the hybrid variable selection and prediction model was used to combine the improved grey correlation analysis method with the Elman neural network to determine the optimal input variables of the single-step prediction model of the indicator variable.In addition,multiple simulation experiments with different input variables and neural network algorithms were designed,and the results showed that the prediction model which combined the improved grey correlation analysis method with the Elman neural network had a higher single-step prediction accuracy for the indicator variable of the air separation system.(3)When performing the multi-step prediction of the industrial process variables,the recursive multi-output method has the characteristics of flexible prediction model,fewer iterations and high calculation efficiency,but there is still the problem of error superposition.Hence a multi-step prediction method based on the recursive multioutput-Markov chain was proposed.This method realizes the correction of the multistep predicted value by calculating the error state of the predicted value,thereby reducing the accumulation of errors in the recursive multi-output prediction process.Additionally,through the example simulation of the air separation system,it is found that the multi-step prediction model of the improved method had smaller prediction errors for the indicator variable,and the prediction trend of the variables is closer to the real situation.