基于混合协同粒子群优化的广义T-S模糊模型训练方法  

Training method for generalized Takagi-Sugeno fuzzy model by hybrid cooperative particle swarm optimization

作　　者：周欣然[1,2] 滕召胜[2] 易钊[2] 

机构地区：[1]中南大学信息科学与工程学院,湖南长沙410075 [2]湖南大学电气与信息工程学院,湖南长沙410082

出　　处：《系统工程与电子技术》2009年第5期1189-1193,共5页Systems Engineering and Electronics

基　　金：国家自然科学基金(60872128);国家技术创新基金(07C26214301740)资助课题

摘　　要：针对广义Takagi-Sugeno(T-S)模糊模型训练中存在的高维、非线性、混合参数估计问题,提出了一种基于混合协同粒子群优化的广义T-S模糊模型训练方法。该方法用离散二进制微粒位置表示模型的结构参数,用普通微粒位置表示模型规则中模糊集隶属函数的参数;这两种微粒位置联合体构成一个模型完整的模型前件参数集。两种群通过协同进化优化所有前件参数;模型后件参数用卡尔曼滤波算法估计。该方法不要任何先验知识,能产生紧凑的、泛化性能较好的模糊模型。函数逼近的数字仿真说明了该方法的有效性。To solve the high-dimensional, nonlinearity, mixed parameter optimization problem during train ing generalized Takagi-Sugeno fuzzy model （GTSFM）, a method for training GTSFM is proposed using hybrid cooperative particle swarm optimization. The structural parameters of models are denoted by the position of discrete binary particles, and the parameters of the membership function in the model rule are denoted by the position of ordinary particles. The combination of positions of the two kinds of particles constitutes a complete premise parameters set of a model. All reasoning parameters are adjusted by cooperative coevolution of two par ticte swarms; all consequent parameters are estimated by Kalman filtering algorithm. The method does not re quest any previous information about objects and is able to produce a compact fuzzy model with the better properties of generalization. The numerical simulation of function approximation shows the validity of the method.

关 键 词：广义Takagi-Sugeno模糊模型 混合协同粒子群优化 协同进化 模型训练 卡尔曼滤波算法 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]