Researches On The Nonlinear System Identification Algorithms Of Hammerstein Model

System identification is playing a very important role in control area and other technical areas, and nonlinear system identification has been a research focus and difficulty. All physical systems are nonlinear to some extent and it is naturally better to use nonlinear model to describe a real system. Nonlinear block oriented model is proved to be effevtive in describing nonlinear systems. The block oriented system is composed of a static nonlinear block and a dynamic linear block, and can be labeled as Hammerstein system or Wiener system depending on the sequence of the connection of the two blocks. This paper is focused on the identification of Hammerstein system. After introducing HARMAX-LSI and HARMAX-RLS algorithm for later use, this thesis proposes three algorithms for identifying Hammerstein system: nonlinear Recursive Instrumental Variable algorithm (RIV), Output Error-PSO algorithm (OE-PSO), Maximum Likelihood-Modified Adaptive Particle Swarmm Optimization (ML-MAPSO), and verifies their effectiveness.Major work and contributions of this thesis are as follows:[1] Nonlinear IV and RIV algorithms for nonlinear system identification are presented, and the case studies based on Hammerstein ARMAX model are carried out. These algorithms are compared with the classic identification method HARMAX-RLS. and the results show that the proposed methods are not only effective but also superior to HARMAX-RLS both in precision and converging speed.[2] PSO algorithm and Output Error method (OE-PSO) are introduced into the identification of Hammerstein system. Without transforming Hammerstein model and identifying extras parameters, OE-PSO algorithm gets the estimated parameters directly. The biggest advantage of OE-PSO is that it is very simple in structure and is easy to carry out. Comparison between OE-PSO and HARMAX-RLS under different noise-to-signal-ratio reveals that the proposed method achieves smaller identification error and the output estimated fits real output better.[3] A Maximum-Likelihood principle and adaptive PSO based nonlinear system identification method are proposed, which uses ML:s excellent asymptotic property and PSO’s strong searching ability to improve identification result. Moreover, a new evolution state estimate mechanism and evolutionary factor are defined to further improve the optimal searching ability of PSO (named MAPSO). Comparisons among HARMAX-RLS, ML-PSO and ML-MAPSO are carried out. With the help of the optimal asymptotic properties of maximum likelihood method and the improved searching ability of MAPSO, ML-MAPSO achieved better result both in precision and converging speed.