Abstract:

In order to solve the problem that the current fuzzy system modeling and optimization methods have not paid enough attention to the number of rules and the structural optimization of the learned fuzzy system, which affects its accuracy and interpretability, a fuzzy system optimization method that combines simulated annealing with support reduction rules is proposed. The method reduces the redundant rules of the system through support calculation and improves the interpretability of the fuzzy system; and uses simulated annealing algorithm to optimize the membership function parameters of the fuzzy system to improve the accuracy of the fuzzy system. For the regression problem, compared with BP（Back Propagation）, RBF（Radial Basis Function） algorithm and classic fuzzy algorithm WM（Wang-Mendel） on three classic data sets in different fields, the experimental results show that the proposed algorithm has achieved higher accuracy in prediction. Compared with WM algorithm, the number of algorithm rules proposed is significantly reduced.

Key words: fuzzy system, simulated annealing algorithm, degree of support, redundancy rule, interpretability

摘要：

从数据中学习模糊系统是其智能建模的重要方法之一，针对目前模糊系统建模及优化方法对于学习后的模糊系统的规则数以及结构优化关注不足而影响了其精度和可解释性的问题，提出了一种结合模拟退火与基于支持度约简规则的模糊系统优化方法。该方法通过支持度约简系统冗余规则进而提高模糊系统的可解释性；同时利用模拟退火算法优化模糊系统的隶属度函数参数进一步提高模糊系统的精度。针对回归任务，与BP（Back Propagation）神经网络、径向基（Radial Basis Function，RBF）神经网络以及经典的模糊算法WM（Wang-Mendel）在不同领域的3个经典数据集上进行实验比较，实验结果表明：该算法在预测方面取得了更高的精度；与WM算法相比，所提算法中规则数明显减少，进一步提高了系统的可解释性。

关键词: 模糊系统, 模拟退火算法, 支持度, 冗余规则, 可解释性