Identification Of Nonlinear Systems Based On A Novel Artificial Bee Colony Algorithm

The problem of identification of nonlinear system has been a hot issue in industrial practice in recent years. Traditional identification methods cannot be applied to nonlinear system directly, such as the least square method (LS), maximum likelihood method, etc. Identification of nonlinear system has important actual meaning and research value. This thesis mainly aims at the identification of Wiener system and Hammerstein system which are typical nonlinear systems. Combined with optimization algorithms proposed in recent years, problems such as slow convergence speed, error accumulation and multivariable system identification are considered.Firstly, the development of optimization algorithm in recent years are summarized. An efficient, simple intelligent search algorithm artificial bee colony algorithm (ABC) is selected through the comparison. By analyzing the working principle of ABC algorithm, a modified artificial bee colony algorithm (NABC) is proposed by combing the advantage of random search, which improves the convergence speed and global search ability of ABC algorithm. Simulations by some classical test functions illustrate the Superiority of the proposed method. NABC algorithm is used as an efficient tool for parameter identification process.Secondly, an identification method of Wiener system based on steady-state responses is detailed, which identifies the static nonlinearity by using step responses and estimates the dynamic linear block by using PBRS signals. To avoid error accumulation, this thesis proposes a two-step estimation method of Wiener model which identifies the linear block based on the limited information of the structure of the nonlinearity including sign information, origin information and monotonic information, but not the estimated parameters. The nonlinearity and the linear block are identified independently by transforming the identification problems to optimization problems. A numerical example and a simulation of pH neutralization process are given to show the effectiveness of the proposed methods.Finally, identification method based on step-responses is extended to multivariable Hammerstein models in this thesis. The estimated internal signals can then be obtained to identify the multivariable linear block. The Sub-Model (SM) method is introduced to identify the linear block. Simulation of a polymerization process is discussed to illustrate the effectiveness of the proposed theory.