The Research And Application Of The Nonnegative Garrote Based Soft Sensor In The Air Separate Process

The oxygen,nitrogen and some rare gases produced by the air separation unit are widely used in metallurgical,chemical,petroleum,mining,military and other sectors.The application of oxygen steelmaking,blast technology and the rapid development of nitrogen fertilizer industry promote the development of large-scale and integrated air separation unit.In order to ensure the quality of the products and improve the safety and reliability of the air separation unit,it is necessary to monitor the oxygen concentration of the production process.The main purpose of this study is to find a soft measurement method to measure the oxygen concentration in real time and keep it in a certain range,so as to overcome the problems that sensors fail to realize the on-line monitoring because of time-delay.Soft sensing technology is a hot topic in the fields of process control.The typical auxiliary variable selection method is the optimal subset method,but with the increase of the data dimension,the computational complexity of the method will increase rapidly and the stability of the selection result is poor.Currently,the main methods for variable selection are based on the penalty function.By compressing the penalty coefficient,this method can not only screen variables but also carry out parameter estimation,such as nonnegative garrote(NNG)and least absolute shrinkage and selection operator(LASSO)etc.In this paper,we develop a new variable selection method for soft sensor applications using the improved NNG.The improved algorithm considers the selection of linear regression model based on the V fold cross validation method,which is to improve the original V cross validation by adding the improved AIC criterion to solve the over fitting problem in the cross validation process when the sample size is large and the data correlation is high.The effectiveness of the algorithm is verified by a simulation example,and compared with the least square method and the traditional NNG algorithm,the simulation results proved the superiority of the algorithm.The proposed algorithm is applied to the oxygen concentration soft measurement of the air separation unit.The results show us that the algorithm can accurately measure the oxygen concentration of the unit,and the precision of the model is better than the least square method and the traditional NNG algorithm.