Selective Ensemble Learning Modeling for Combustion Process of Coke Oven

In view of the complicated and non-linear combustion process of coke oven, this paper adopts selective ensemble learning method to establish the model of combustion process of coke over. Firstly, the data of the combustion process are analyzed by gray correlation analysis to determine the model input variables. Then, Bootstrap sampling is used to generate several different sub-training sets from the actual operation data of coke oven. Based on each sub-training set, LS-SVM is used to model. Furthermore, the optimal parameters are automatically determined by web search method. Finally, a method that based on the error complementarity of learning machine to solve the diversity value of base learner set is studied. By using this method, the redundant base learner is deleted and the ensemble learning model is determined. Matlab simulation test proves the feasibility and effectiveness of the scheme.