基于遗传优化HMM的模拟电路早期故障识别和诊断

Recognition and diagnosis of incipient faults in analog circuit using genetic optimized HMM

针对Baum-Welch(B-W)算法易陷入局部最优解的问题,采用多智能体遗传算法对隐马尔可夫模型(HMM)进行参数优化估计,设计了染色体编码结构和遗传操作方式。为了使HMM更适合于模拟电路的故障识别和诊断,将状态转移概率矩阵改为时变矩阵,利用监测数据,通过多智能体遗传算法实现状态转移概率的更新。将改进后的HMM用于模拟电路早期故障的识别和诊断中,并采用线性判别分析(LDA)方法对测量信号进行特征提取。仿真结果表明,改进后的HMM具有更强的故障识别和诊断能力。

Multi-Agent Genetic Algorithm （MAGA） is used to optimize and estimate parameters of Hidden Markov Model （HMM） for overcoming the deficiency that Baum-Welch algorithm is easy to fall into local optimal solution. Chromosome coding structure and genetie operation mode were designed. In order to make the HMM more suitable for analog circuit fault recognition and diagnosis, the state transition matrix was improved to time-varying one that was updated by Multi-Agent Genetic Algorithm （MAGA） according to the monitoring data. This improved HMM was applied to recognize and diagnose the incipient faults of analog circuit, and the Linear Diserinlinant Analysis （LDA） was used to reduce dimensionality and remove redundancy of the voltage feature vectors. The experimental results indicate that the improved HMM has the better fault recognition capability than the traditional HMM.