An Optimal Regions-based Multiple Model Identification Approach For Nonlinear Systems

With the increasing complexity of the industrial process,most control systems are multi variable,nonlinear,and have a wide operating range.In order to improve the control performance of complex dynamic systems,the identification and control methods of nonlinear systems based on multiple models are more and more concerned.In the present paper,the identification methods of one-dimensional scheduling variables and multidimensional scheduling variables that based on optimal boundary partition are proposed for the nonlinear dynamic system.The corresponding relationship between sample points and submodels is highlighted,and the advantages of the identification model are illustrated from aspects of model precision and control performance.The main achievements of this paper are as follows:1)In order to solve the problem that scheduling variables are one-dimensional,an optimal boundary partition method is proposed to minimize the output error.This method takes full account of the influence of the submodels' boundary on the accuracy of the model.Submodel parameters are identified on the basis of precise data partition.Compared with the identification model based on clustering method and the linearized model based on the operating point,the proposed method can effectively improve the accuracy of the model under the same scheduling variables.2)For the case that the scheduling variables are multidimensional,an optimal boundary partition method is proposed by using the parameter vector clustering and the Softmax classification.The method can solve the problem that the boundary of submodels is not easy to be initialized and the submodels' operation regions are not completely divided under the multi-dimensional scheduling variables,so that the identification method based on the optimal boundary partition can be popularized.3)Based on the obtained PWA model,a global MPC controller based on MLD framework is designed.By comparing the control effects of different identification models under the same predictive controller,the superiority of the proposed identification method in the control performance is further verified.