Research On Prediction Of Oxygen Production In Air Separation Unit Based On Machine Learning

With the the revolution of industrial intelligence,the industrial big data technology has been widely used.For the current problem of high resource cosumption and heavy pollution in iron and steel industry,Using the big data technology to improve the production efficiency and reduce polluion is of great significance in research and application.The Air separation unit(ASU)for Oxygen is one of the important part in iron and steel industry.By exploiting the data mining technology and machine learning technology to analyze the historical data of the ASU and establish the prediction model of oxygen production,which solves the long-standing imbalance between oxygen supply and demand,and provides guidence for the scheduling in the enterprise.In this paper,a series of data-driven oxygen production prediction models are proposed based on the historical data of the ASU in an iron and steel enterprise.we preprocessed the raw data with abnormal detection and standardization to improve the quality of the data,then we select features by using Lasso method to reduce the dimensionality of data in preparation for the modeling.Secondly,Based on the pre-processed data set of ASU,three algorithms,including Random Forest,BP neural network and LSTM neural network,were used to established the prediction models of oxygen production.The experimental results showed that the LSTM neural network has the best performance among these three individual models,which had the lowest Root Mean Squared Error(RMSE)and Mean Absolute Percentage Error(MAPE)on the test set.Finally,Based on the individual models,we designed two ensemble prediction models by using average weighting method and Stacking method.The results show that the two ensemble models is better than three individual models in accuracy,and Stacking prediction model has the better performance,Therefore,Stacking prediction model is suitable for predicting the oxygen production of the ASU.