Research Of Blocked-Oriented Nonlinear System Identification Based On Intelligent Computation

Nonlinear system identification has always been one of the key problems in process control.As a kind of nonlinear model,the blocked-oriented model is simple in structure,easy to operate,and can effectively fit the actual production system,so it is often used in the identification of nonlinear system.Considering the rapid development of intelligent computing in recent years and the successful solution of many nonlinear system identification problems,this paper mainly focuses on two typical blocked models:Wiener model and Hammerstein model,and studies specific identification schemes by combining intelligent optimization algorithm and neural network.The main contents include:(1)Aiming at the shortcomings of PSO,such as poor local search capacity and easy premature convergence,a modified particle swarm optimization algorithm was proposed by introducing premature convergence evaluation indicators and adaptively adjusting the position of the population after classification(MPSO).Aiming at the inefficiency of the horizontal positioning mechanism of the standard moth optimization algorithm(MFO),the idea of Gaussian mixture distribution is introduced into its position update and population initialization.The test function proves its effectiveness.(2)The Wiener model is studied.The linear module is represented by dynamic linear neurons,and the nonlinear module is approximated by BP neural network and RBF neural network,respectively.The model is expressed as a series network structure.The bilevel optimization strategy is applied to the identification.BP algorithm is used in the inner layer learning,while MPSO algorithm is used in the outer layer learning.After that,the effectiveness of the identification scheme is proved by the identification of CO2 concentration system.(3)Study the single-input single-output Hammerstein model.The nonlinear block is represented by Function link Artificial Neural Network(FLANN).By setting the mean square error(MSE)as an evaluation index,the proposed GMFO algorithm is used to train the FLANN-IIR structure.It combines the computing power of GMFO algorithm and the nonlinear fitting ability of FLANN well.Finally,simulation examples prove the effectiveness of the identification scheme.(4)Aiming at the identification problem of multiple input multiple output Hammerstein model under the influence of heavy-tailed noise.Considering that the traditional identification algorithm has a large error or even fails under the interference of heavy-tailed noise,and the Gaussian-mixture flight strategy in the GMFO algorithm can effectively deal with the interference of outliers in heavy-tailed noise,we use RBF neural network to fit the static nonlinear block,and the GMFO algorithm is used to synchronize the training of RBFNN and the parameter identification of the linear part.Finally,simulation experiments prove the effectiveness of the identification scheme and its robustness to outliers.