Development Of Parameter Identification For Wiener Model

Wiener model is a block-oriented nonlinear system consisting of a cascade of dynamic linear block and static nonlinear block.Due to its simple structure and good approximation performance,Wiener model is widely used in industrial process control,such as pH neutralization process,distillation column,mechanical system,and communication system.However,there still exists some challenges in the parameter identification of Wiener model,such as the location of noise and the structure identification of nonlinear system.Therefore,this paper conducts an in-depth study on the parameter identification of Wiener model.The specific work is as follows:A modified Brain Storm Optimization(mBSO)algorithm is proposed for the parameter identification of Wiener model with output noise when the structure of model is known.Firstly,based on the classical Brain Storm Optimization algorithm,diverse combination of individuals and the introduction of a dynamic adjustment factor are introduced.Then the searching domain of individuals can be adaptively adjusted and the speed of mBSO can achieve quickly convergence.In this paper,the Markov model of mBSO algorithm is used to prove the convergence of proposed algorithm.The effectiveness of mBSO algorithm is verified by numerical simulation and a CE8 couple electric drive system.A Cubic Spline Approximation-Bayesian Composite Quantile Regression(CSABCQR)algorithm is adopted for the parameter identification of Wiener model with process noise when the structure of model is unknown.The high-order ARX model is utilized to represent the linear block,while the nonlinear block is approximated by cubic spline function,assuming that the nonlinear part is reversible.Based on Bayesian principle,the Bayesian Composite Quantile Regression algorithm is applied to identify the parameters.In order to reduce the amount of calculation,the Markov Chain Monte Carlo algorithm is introduced to calculate the expectation value of posterior distribution,that is,the parameter to be estimated.In the decision of structure order,we choose Final Output Error criterion determine the nonlinear order,and select AIC method for linear block.Finally,a numerical and an industrial simulation verify the effectiveness of the algorithm.In this paper,two different parameter identification methods are proposed for Wiener models with various noise source locations.When the noise is at the output of the whole system,we choose mBSO algorithm.When it is located in the system,we choose CSA-BCQR algorithm.The experimental results showed that both algorithms could obtain better performance in convergence speed and identification accuracy compared with other algorithms.