一种基于主元递归分析法的多模糊逻辑系统的组合形式

Modular Fuzzy Logic System Based on Principal Component Regression Analysis

根据多个模型相加可以提高整体预测精度和鲁棒性的思想,提出了一种具有递阶特点的模糊逻辑模型.该模型采用基于山峰函数的减法聚类算法,将样本数据集分成多组来进行局部模糊模型的建立和训练,大大提高了组合模糊逻辑模型的训练效率.各局部模糊系统的预测输出通过主元递归分析法(PCR)连接,解决了模型之间的严重相关性问题,增强了模型的预测能力,提高了模型的鲁棒性.仿真结果表明,组合多个模糊逻辑模型能够达到比局部模型更好的建模效果,并能有效地改善模型的预测能力和泛化能力.

According to the idea that combined models can improve accuracy and robustness, a modular fuzzy logic model is brought forward. First, the subtraction clustering method based on the mountain function is used to obtain group data such that training efficiency is greatly improved. Then, the local fuzzy logic models are respectively set up. Finally, principal component regression (PCR) analysis method is adopted to get the forecasting output of the whole model. Thus, the severe correlation among models is avoided, and the precision and robustness is greatly improved. The proposed system has been evaluated by a nonlinear function. The simulation results demonstrate its better precision, feasibility and robustness than those of the local model.