摘要
提出了一种基于遗传算法、神经网络、模糊集理论与数据融合技术相结合的模拟电路故障诊断新方法。该法使用多类电路测试数据来解决由于测试节点不足而带来的故障信息欠缺等问题,采用遗传算法来优化BP网络的结构与初始权值分布。对每类测试信息各用一个独立的所提遗传神经网络进行初步诊断,得到基于各类测试信息的被诊断电路属于不同故障状态的可能性。在充分考虑每个神经网络输出信息重要程度的基础上,采用模糊积分融合方法进行决策融合。文中研究了其故障特征提取、样本选择、诊断系统结构、学习算法及其综合决策方法,并通过电路诊断实例,阐述了该方法的具体实现,验证了所提方法的鲁棒性。诊断结果表明:所提方法对容差模拟电路的多故障与单故障诊断均适用,故障定位准确率高。
Based on genetic algorithm（GA）, neural network, fuzzy theory and data fusion technology, a new fault diagnosis method for analog circuits with tolerance is proposed The feature extraction, selection of training samples, structure of diagnosis system, GA-BP algorithm and synthetic decision method are dealt with. The proposed approach selects multiform circuit signatures to diagnose the circuits without sufficient accessible nodes and uses GA to optimize the structure and original weight distribution of BP networks. The GA-BP networks are adopted to implement the local diagnosis and each of them classes faults based on single kind of circuit signatures. Under considering the importance of output information from the GA-BP networks, the decision fusion is performed by using fuzzy integral. The realization of the proposed strategy is expounded by using a practical circuit. The experiment results show that the proposed method has the capability to diagnose multiple faults and single faults in tolerance circuits and gains satisfactory accuracy.
出处
《中国电机工程学报》
EI
CSCD
北大核心
2006年第3期19-24,共6页
Proceedings of the CSEE
基金
国家自然科学基金项目(60375001/F0304)
国家科技攻关计划项目(2003BA104C)
湖南省自然科学基金项目(04JJ6034)
教育部新世纪优秀人才支持计划(NCET-04-0767)
湖南省科技计划项目(04FJ2003
05FJ3008)~~
