The Study Of Neural Network Ensemble And Its Application In PTA Process Industry

Recently, neural network ensemble has become one of hot topics of machine learning technology. And it has the advantages of generalization ability and stability than a single artificial neural network. Selective ensemble technology received extensive attentions because of its advantages of applicability and combinability.Purified terephthalic acid (PTA), as a major chemical raw material, covering all aspects of the national economy, has been applied widely in various fields. Thus, the study of PTA production process has great practical significance. In order to solve the lack of generalization and stability in single neural network model because of the complexity of mechanism and operating point drift, the paper puts forward a new selective neural network ensemble approach named DWSEN which to measure the diversity of individuals according to weights of individual networks. And it presents the formula used to calculate the diversity of networks with same network structure. Comparing with some prevailing ensemble approaches such as Bagging and Boosting, experiments reflect that DWSEN has higher generalization ability and stronger stability. The method is applied to the modeling of evaporation of Film Evaporator and tower conductivity of Solvent Dehydration Tower in PTA industrial process. Cases study show that the obtained models can possess better generalization performance, and simulate the production process better in PTA industry. DWSEN provides a well-estabilished modeling program for complex mechanism model in PTA process industry.