一种基于改进客观聚类分析的模糊辨识方法

A fuzzy identification method based on the enhanced objective cluster analysis

针对模糊辨识中采用迭代和人为决策法确定模糊规则数时易受噪声和人为因素的影响,而导致算法鲁棒性较差和计算量较高的问题,提出一种基于改进客观聚类分析的模糊辨识方法.首先引入并改进了客观聚类分析法,克服了迭代导致的规则数冗余,降低了人为因素对聚类结果的影响,从而减小了计算量并提高了鲁棒性;然后结合模糊聚类和稳态卡尔曼滤波法,分别辨识了前提和结论参数;最后通过Box-Jenkins仿真实例验证了所提方法的有效性.

In fuzzy identification, iterations or human decision making are usually used to identify fuzzy rules. However, the clustering result is possibly affected by noise and artificial factor, which results in weak robustness and high computation cost. In this paper, a fuzzy identification method based on the enhanced objective cluster analysis is presented. Firstly, the objective cluster analysis algorithm is introduced and enhanced such that the redundant rule numbers caused by iterations is overcomed and the effect of human factor on the clustering result is decreased. Therefore, the computation burden is reduced, and the robustness of the algorithm is improved. Then, the premise parameters and the consequence parameters are identified by fuzzy c-means clustering algorithms and the stable Kalman filter algorithm respectively. The effectiveness of the proposed method is verified by the example of BoxJenkins gas furnace simulation.