Research On Intelligent Computer Algorithm For Hammerstein Model Identification Under Heavy Tail Noise

For the modular Hammerstein model,the least square method is often used to identify the model under the assumption of Gaussian noise,and the complex disturbances such as large-scale outliers are not considered enough,which results in identification difficulties.Such disturbances show the characteristics of "peak and heavy tail" in the distribution pattern,and can be described by heavy tail noise.The parameter identification of Hammerstein model under heavy tail noise has become an important research direction.Intelligent computing method,as a new computing form,has achieved good results in the field of system identification.It can effectively solve the problem of noise processing which is difficult to solve by traditional identification methods.Therefore,this paper introduces intelligent computing to the identification of Hammerstein model parameters under heavy tail noise.The details are as follows:Firstly,aiming at SISO Hammerstein model,a network Hammerstein model structure is proposed by using the non-linear module of BP neural network fitting model,and the learning rule of error back-propagation based on gradient method is deduced.Aiming at this model,a parameter identification method of two-level optimization strategy is proposed,and cuckoo algorithm with random flight strategy is used as outer loop.In order to weaken the influence of heavy-tailed noise,the back-propagation of BP idea is used as an inner loop to optimize the model parameters.The effectiveness of the proposed method is verified by the identification experiment of heavy-tailed noise with mixed Gauss distribution.Then,aiming at the MIMO Hammerstein model,using Taylor series to approximate the non-linear part of the model,a random search cuckoo algorithm based on fuzzy logic is proposed.By designing the fuzzy logic,the non-linear random search algorithm is introduced into the cuckoo search algorithm.At the same time,the output of the fuzzy logic is used to improve the search step of the cuckoo algorithm,Levy flight strategy and non-linear flight strategy are used.Sex random search strategy is used to weaken the influence of heavy-tailed noise on system identification.The identification experiments with heavy-tailed noise with mixture Gaussian distribution and heavy-tailed noise with t distribution verify the effectiveness of the algorithm.In addition,compared with the traditional Hammerstein model parameter identification method,the proposed two-level learning strategy and parameter optimization algorithm based on intelligent computing can better deal with the parameter identification problem of Hammerstein model under heavy tail noise.