The Research On Algorithm Of Designing And Optimizing T-S Fuzzy Controller

An approach for designing and optimizing Takagi-Sugeno type fuzzy logic controller is proposed in this paper, which is based on experience data. The algorithm has been successfully verified on inverted pendulum. Different with the common method of designing T-S type fuzzy logic controller based on T-S model, T-S type fuzzy logic controller is directly constructed and optimized from the expert's experience data set in the algorithm. The whole algorithm is divided into three stages. The first algorithm is made up of three steps. Firstly, the Gaussian membership functions of input variables are determined according to the range of input variables by average division. Then, the coefficients of T-S type fuzzy logic control rule consequent are determined with recursive least square algorithm while the parameters of membership function are fixed. Finally, all the parameters of T-S type fuzzy logic controller, which are made up of antecedent parameters and consequent parameters, are simultaneously optimized with gradient descent algorithm. The second algorithm is also made up of three steps. The first step is to apply the subtractive clustering algorithm in experience data set to determine the clustering centers (namely control rules). The second step is to determine the Gaussian type memberships of input variables according to clustering centers. The third step is to determine the coefficients of T-S type fuzzy logic rule consequent. The first algorithm and the second one are integrated into the third algorithm which owns more advantages. Firstly, the subtractive clustering algorithm is applied in experience data set to determine the clustering centers. Secondly, the Gaussian type memberships of input variables are determined according to clustering centers. Then, the coefficients of consequent are identified with recursive least square algorithm. Finally, the parameters of rule antecedent and consequent are simultaneously optimized with gradient descent algorithm. The effectiveness of algorithms has been verified on inverted pendulum.