Study On Optimization Algorithms For Fuzzy Modeling And Its Applications

The powerful approximation ability of fuzzy model provides an effective modelingmethod for nonlinear and uncertain systems. As a result of its limitations, the traditionalfuzzy modeling methods sometimes cannot meet the requirements of modeling accuracy.So the proposing of intelligent optimization algorithms provides a new way to solve thisproblem. The method of combining intelligent algorithms with fuzzy identification theory,and applying the intelligent optimization algorithms to optimize the antecedent andconsequent parameters of fuzzy model is an effective way to improve the modelingaccuracy and achieve satisfactory identification result. This paper mainly studies theparameters optimization of fuzzy model based on intelligent optimization algorithms, themain study content states as follows:Firstly, the research significance of this paper is given, and the development processand research status of the intelligent optimization algorithms and T-S fuzzy model aresummarized. In addition, the common intelligent optimization algorithms used in fuzzyidentification are introduced, and the simulation results of parameters optimization basedon different intelligent optimization algorithms are compared and analyzed, whichprovides some guidance for further research.Secondly, an improved bacterial foraging optimization (IBFO) algorithm is proposed,which improves the convergence accuracy of BFO algorithm and avoids it stucking intolocal optimum. To the problem of nonlinear systems identification, IBFO is used to updatethe parameters of Gaussian membership functions, and the consequent parameters areupdated by the recursive least square (RLS) algorithm. In this way, the global optimizationof fuzzy model parameters is realized.Then, to the problem of prediction of Mackey-Glass chaotic time series, a robustlearning method based on modified fruit fly optimization algorithm (MFOA) and leastwilcoxon (LW) method is proposed to training T-S fuzzy model. The MFOA is used tooptimize the parameters of Gaussian membership functions, which can improve theidentification accuracy and convergency speed. The LW method is used to update the consequent parameters, when outliers occur in the training data, the strong robustness ofLW to outliers is effective to improve the sensitivity of traditional least mean squaremethod.Finally, this paper proposes an adaptive inverse control method based on fuzzyinverse model identified by PSO algorithm. In the first, the T-S fuzzy model and theGaussian membership functions with uncertain center and width are used, and the fuzzyinverse model is obtained by using PSO algorithm to optimize both of the antecedent andconsequent parameters in the process of inverse modeling in off-line manner. Afterwardsthe initial inverse model is connected to the plant in series as the initial controller and theconsequent parameters of the inverse model are tuned by the variable-step least meansquare (VSLMS) algorithm in on-line manner while its copy is connected to the plant assystem controller. This approach realized the adaptive inverse control of nonlinear system.