Research On Local Model Network Identification And Predictive Control

Industrial process system with multi-variable,highly non-linear and continuous change of working point.How to design a control strategy has always been a hot issue in the industry.Multi-model control based on "decomposition-synthesis" strategy provides an effective strategy for solving such problems.In this paper,local model network identification and predictive controller are studied.The main work includes:In order to model the local model network of the non-linear system,this paper uses GK clustering algorithm to divide the input and output data offline collected by the system into the whole working conditions.The clustering update rule is reconstructed based on Kohonen network idea.And an improved GK clustering algorithm is proposed.On this basis,the global model is decomposed into two "virtual" subsystems,and a two-level iterative least squares identification is given.The simulation examples demonstrate the effectiveness of the algorithm and the accuracy of the LMN model for approximating nonlinear systems.The predictive controller is designed by transforming the LMN model into a state space equation.However,at each sampling time,it is necessary to scroll on-line to solve the optimal value under constraints,which requires a large amount of on-line calculation.Therefore,a heuristic SQP optimization method is proposed in this paper.At the same time,the Laguerre candidate function with appropriate order is used to compress the number of optimization variables,and reconstruct the MPC algorithm.The asymptotic stability of the controller is proved.In this paper,the superiority of predictive controller is verified by step disturbance and slope tracking change of fuel power system.