Research On Nonlinear System Identification Based On Heavy Tailed Distribution Cuckoo Search

After a detailed analysis of the characteristics of the cuckoo algorithm(CS),a series of improvements are made to the cuckoo algorithm(CS)based on the relevant application background,and the effectiveness of the improved CS is proved by the simulation experiment.The improved algorithm is applied to the identification of nonlinear Hammerstein models.The specific contents are as follows:(1)First,we introduced the rapid development of cuckoo algorithm(CS)in recent ten years.Cuckoo algorithm(CS)is a new swarm intelligence optimization algorithm.By analyzing the working principle and search mechanism of cuckoo algorithm(CS).In this paper,according to the different system identification requirements,two improved cuckoo algorithm TTCS and GMDA are proposed based on the different system identification requirements.It can be used as a powerful tool for identification of corresponding system parameters.(2)Nonlinear Hammerstein model for single input and single output is studied.Most of the previous studies of nonlinear Hammerstein models are based on analytical methods,which are difficult to study.Especially,it is difficult to extract analytical solutions from the nonlinear parts themselves.In order to solve this problem,this paper attempts to use the modular nonlinear system identification method to approximate the nonlinear part of the Hammerstein model by using the function connected neural network(FLANN).Accordingly,in order to identify the modularized model parameters proposed above,this paper first proposes to improve the t distribution of a class of typical heavy tailed students and the sequence improved cuckoo algorithm(TTCS)from the interval(0,1)between the students' t distribution.The simulation example of the third chapter confirms the effectiveness of the TTCS algorithm in processing the parameter identification of the single input and single output Hammerstein model.(3)The identification of Hammerstein model for multiple input multiple output systems under heavy tailed noise is studied.Most of the system identification problems in the past are studied under the hypothesis of white noise and white noise based colored noise.In recent years,it is found that noise is assumed to be Gauss noise in some complex industrial problems.However,there is no unified analytical method for nonlinear system identification under such a non Gauss noise.In this paper,we try to adopt modular system identification method,and use radial basis function(RBF)neural network to transform this identification problem into a parameter optimization problem.We use the improved cuckoo algorithm(GMDA)first proposed in this paper to solve the above optimization problem.The simulation example of the fourth chapter confirms the effectiveness of GMDA in solving such problems.(4)We explored the possibility of using artificial intelligence to train neural network.In(2)and(3),the training of neural network(FLANN)and radial basis(RBF)neural network is not used as a traditional gradient descent algorithm,but the training of neural network is transformed into a parameter optimization problem,and the relevant parameters that need to be given by artificial experience are also converted to the values that can be trained by intelligent optimization algorithm.It reduces the difficulty of neural network training and shortens the training complexity.