中文摘要：

连续高斯混合密度隐马尔可夫模型（Continuous Gaussian Mixture Hidden Markov Model, CGHMM）在故障诊断领域得到了广泛应用,取得了较好效果.CGHMM训练模型较大、局部最优,但模型参数初始化值会直接影响迭代收敛速度和模型效用.全局最优的遗传算法（Genetic Algorithm, GA）初始化CGHMM模型参数,为CGHMM训练提供了一个好的初始值,不仅可以加快收敛速度,还可以得到一个更好的模型.通过GA初始化CGHMM、CGHMM训练和CGHMM诊断过程等三个方面的仿真实验和比较分析可以得出,该方法具有训练速度快和CGHMM模型好的优点.在最后的CGHMM诊断仿真实验中,该方法诊断精度为100 %,高于经典方法的96 %,表明GA确实可以成功应用于CGHMM参数初始化,是一种可行的故障诊断方法.

英文摘要：

The continuous Gaussian mixture hidden Markov model (CGHMM) has been widely and successfully used in fault diagnosis. However, the traditional CGHMM has some inherent disadvantages, such as the model complexity, the local optimization, low iterative convergence speed and modeling effect due to initialization of CGHMM parameters. In this paper, the CGHMM model parameters initialized by genetic algorithm were used as the reasonable initial values for CGHMM training. Using this method, the convergence speed can be accelerated and a better effect of modeling can be obtained. Through the simulation experiments in three aspects of CGHMM initialized by genetic algorithm, the CGHMM training process and the CGHMM diagnosis process, it was verified that this method have the advantages of fast training speed and better CGHMM model. The CGHMM diagnosis result demonstrates that the diagnosis precision can achieve 100%, which is higher than that of 96%of the classical method. This result shows that the genetic algorithm can be applied to CGHMM parameter initialization, and the proposed method is a feasible method for fault diagnosis.