Study On Differential Evolution And Its Application In Parameter Identification Of Nonlinear System

System identification is an important content of modern control theory, and nonlinearsystem identification is one of the important branches. Usually, system identification problemcan be transformed into a general optimization problem to solve. Therefore, the optimizationmethod cloud be used to identify system parameters. Traditional identification methods have abetter recognition performance in the field of linear system identification, however, they oftencan not have satisfactory identification results in the nonlinear system identification. In recentyears, some intelligent optimization algorithms such as genetic algorithms, neural networks,particle swarm optimization algorithm is applied to system identification widely, and itopened up new avenues for solving the nonlinear system identification. This paper studies theDifferential Evolution (DE) algorithm and its improvement, and do some works aboutimproved algorithms for parameter identification of nonlinear systems.Firstly, the DE is introduced in detail, including the fundamental and algorithmprocedure. Analyze the influence of control parameters for algorithm and the disadvantage ofthe DE, for the shortcoming of algorithm, processed two improved algorithms, one isMutation Different Evolution (MDE), the other one is Crossover Different Evolution (CDE).The experiments researching were done by four classic testing functions. The experimentresults show that the improved DE has high optimization ability, high accuracy and highconvergence. Compared to the basic DE, the improved DE are much more feasibility andeffectiveness.Secondly, the two improved algorithm is applied to two kinds of system identification.MDE is used to identify the parameters of Hammerstein model, and CDE is used to identifythe parameters of Wiener model. After several experiments, the results show that MDE hasglobal search capability, high recognition accuracy; CDE has local search ability and fastconvergence. Both the two improved algorithm can identify the model parameters quickly andaccurately, and establish reliable models.Finally, in order to improve the accuracy and convergence speed of the algorithm,proposed a new improved algorithm. Adjusting the crossover probability can speed up theconvergence rate. Two typical nonlinear models are tested, the result shows that the algorithmcan effectively avoid premature convergence, significantly improve the ability of globalsearch algorithms. In order to verify the practicality of the algorithm, taking Glutamic acid forthe study, the improved algorithm is applied to the growth model of Glutamic acidfermentation. Compared the data of the experiment with other algorithms, the results provethat the proposed algorithm has improved robustness and faster convergence rate.