A Systematic Approach in the Identification of Fuzzy Models

• Main Authors: Chen, Yi-Ru, 陳怡儒
• Other Authors: Yo-Ping Huang
• Format: Others
• Language: zh-TW
• Published: 1997
• Online Access: http://ndltd.ncl.edu.tw/handle/58153930758375628240

碩士 === 大同工學院 === 資訊工程學系 === 85 === The purpose of this thesis is to propose a systematic and effective method toconstruct different types of fuzzy models such as Tsukamoto-type, Mamdani-type,and Takagi-Sugeno-Kang-type fuzzy models. The main processes of the proposedsystematic fuzzy modeling scheme are as follows. The first step is to fuzzycluster the given data and then to find the cluster centers. However, thetraditional clustering methods such as fuzzy c-means algorithm and mountainclustering method, when the given data are too concentrated on a compact area,the cluster centers cannot be well estimated due to the mutual effect ofneighboring data. Therefore, a modified clustering algorithm is developed toovercome this drawback. In the developed algorithm, the center of the mostsatisfied subset is assigned to the considered cluster center after tuningthe centers several times. Later, the subset with the center just identified istemporarily removed to eliminate the effects of this cluster to the others. Bymeans of this scheme, we can achieve not only the goal of fast clusterung butalso good clustering performance. The second step is to establish a fuzzy rulebase. The centers of the membership functions are obtained by the projectionof the cluster centers onto the axes of their own coordinate. By comparingthe maching degrees, we can gradually build the necessary fuzzy rules for therule base. The final step is to tune the widths and the centers of the triangular membership functions via the genetic algorithms and the gradientdescent method. Since the center of area defuzzification method is not allowedto take the partial derivatives, the parameters of such membership functionscannot be tuned by the gradient descent method. Therefore, the defuzzificationof the Mamdani-type fuzzy model is reformulated to solve the partial derivativeproblem. Based on the proposed method, we perform simulations for various commonly used examples. The simulation results are given. Based on the simulation results, we can find that the proposed fuzzy modeling approachesprovides very satisfactory performance.