基于改进离散粒子群优化算法的容差电路故障特征提取  

Feature Extraction Method for Tolerance Circuit Fault Diagnosis Based on Improved Basic Particle Swarm Optimization Algorithm

作　　者：刘红[1] 曹颖[1] 隆腾舞 

机构地区：[1]长春理工大学光电工程学院,吉林长春130022

出　　处：《兵工学报》2015年第8期1494-1501,共8页Acta Armamentarii

基　　金：吉林省自然科学基金项目(201115160)

摘　　要：采用故障信息量对容差电路输出信号中的故障征兆进行描述,采用等间隔选取特征点、单特征点诊断信息量最大和多特征点联合诊断信息量最大3种不同的特征子集选取规则,提出了基于改进映射函数、自适应权重、基于自然选择以及基于自然选择和自适应权重的4种离散粒子群优化(BPSO)算法对特征子集进行搜索的方法,并将获取的不同最佳特征子集分别用于训练不同的神经网络,并用训练好的神经网络完成容差电路的故障定位。仿真实验结果证明了容差电路故障特征子集的改进BPSO搜索算法的有效性,故障定位效率可达95.2%.The fault information entropy is used to describe the fault symptoms of output signal of tolerance circuit. Three different feature subset selection rules are adopted, such as equal interval-selected feature point, and feature points selected by maximum information entropy of single feature points and joint information entropy of multiple feature points. Four kinds of improved basic particle swarm optimiza- tion （BPSO algorithms are proposed to search the fault feature subsets. These four algorithms are im- proved mapping function BPSO algorithm, adaptive weighting BPSO algorithm, natural selection-based BPSO algorithm, and BPSO algorithm based on natural selection and adaptive weighting. The optimal fea- ture subsets obtained by feature extraction are used to train the neural networks as classifier. The fault lo- cation of tolerance circuit is completed using a trained neural network . Experimental results show that the optimal feature subset searching methods based on improved BPSO algorithm are valid, and the accuracy of fault location can reach 95.2%.

关 键 词：信息处理技术 信息熵 特征提取 改进离散粒子群优化算法 容差电路 

分 类 号：TN707[电子电信—电路与系统]