基于改进PSO与规则约简的模糊系统优化算法  

Fuzzy System Optimization Algorithm Based on Improved PSO and Rule Reduction

作　　者：蔡际杰 陈德旺 黄允浒 黄玮 CAI Jijie;CHEN Dewang;HUANG Yunhu;HUANG Wei(College of Mathematics and Computer Science,Fuzhou University,Fuzhou 350108;Key Laboratory of Intelligent Metro of Universities in Fujian Province,Fuzhou University,Fuzhou 350108;School of Computing and Information Science,Fuzhou Institute of Technology,Fuzhou 350506)

机构地区：[1]福州大学数学与计算机科学学院,福州350108 [2]福州大学智慧地铁福建省高校重点实验室,福州350108 [3]福州理工学院计算与信息科学学院,福州350506

出　　处：《计算机与数字工程》2021年第8期1525-1530,共6页Computer & Digital Engineering

基　　金：国家自然科学基金面上项目(编号:61976055);智慧地铁福建省高校重点实验室建设基金项目(编号:53001703,50013203)资助。

摘　　要：模糊系统是一种具有强可解释性和高鲁棒性的智能方法,但目前仍存在精度不高、产生的模糊规则太多等缺陷。针对目前存在的问题,论文通过改进粒子群优化算法优化模糊系统高斯型隶属度函数的参数,以及计算规则支持度约简模糊规则,提出了CPSFS和SPSFS两种模糊系统优化算法。在两个不同领域的经典数据集上的研究结果表明:1)CPSFS算法在训练集和测试集上的预测精度明显优于传统的BP神经网络、RBF神经网络、线性回归等算法;2)CPSFS算法与SPSFS算法减少了大量模糊规则,保证了模型的可解释性;3)CPSFS算法在约简模糊规则后预测精度依然表现最优,符合新时代下回归问题对于AI技术的要求。Fuzzy system is a kind of intelligent method with strong interpretability and high robustness,but at present,there are still some defects,such as low precision,too many fuzzy rules and so on.Aiming at the existing problems,this paper proposes two fuzzy system optimization algorithms,which are CPSFS and SPSFS by improving particle swarm optimization algorithm to optimize the parameters of gauss membership function of fuzzy system,and reducing the fuzzy rules by calculating the support degree of rules.The research results on two classical data sets in different fields show that the prediction accuracy of CPSFS algorithm in training set and test set is obviously better than that of traditional BP neural network,RBF neural network and linear regression algorithm.CPSFS algorithm and SPSFS algorithm reduce a lot of fuzzy rules and ensure the interpretability of the model.CPSFS algorithm still has the best prediction accuracy after reducing the fuzzy rules,which meets the requirements of AI technology for regression problems in the new era.

关 键 词：模糊系统 可解释性 鲁棒性 粒子群优化算法 高斯型隶属度函数 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]