采用自适应基因粒子群算法优化隐马尔科夫模型的方法及应用

Adaptive Genetic Particle Swarm Algorithm for Optimization Hidden Markov Models with Applications

针对隐马尔科夫模型参数学习算法易收敛于局部极值的问题,提出了一种自适应基因粒子群算法,并将该方法应用于隐马尔科夫模型的训练,实现对隐马尔科夫模型初始参数的优化。在基因粒子群算法的原理以及操作流程的基础上,采用了自适应的参数调整方法,提高了基因粒子群算法的优化性能。分析了所提方法的全局、局部搜索能力以及收敛速度,开展了不同状态滚动轴承的故障诊断实验和测试,并与基于粒子群算法优化隐马尔科夫模型初始参数的方法进行对比。实验结果表明,所提方法对正常、内圈故障、外圈故障以及滚动体故障轴承的诊断准确率均能达到100%,相比于基于粒子群算法优化隐马尔科夫模型初始参数的方法,最高将分类正确率提高了28.57%、分类离散度提高了268.58%,证明了方法的有效性和准确性。

Aiming at the problem that the parameters learning algorithm of hidden Markov model easily converges to local optimal solutions, an adaptive genetic particle swarm algorithm is proposed,which is applied to the parameters learning algorithm of hidden Markov model. The initial parameters of hidden Markov models are optimized,and the principle and process of genetic particle swarm optimization algorithm are introduced. The adaptive method is adopted tthe performance of genetic particle swarm optimization algorithm. Global,local search capability and convergence rate of the proposed method are analyzed. Experiments and tests of different bearing conditions are carried out,and vibration signals are collected. The results show tcorrect classification rate of the proposed method for the bearing with normal state, inner ringfault,outer ring fautt or rolling element fault reaches 100%. Compared with the method ofoptimizing the initial parameters of hidden Markov model based on particle swarm algorithm,the correct classification rate of the proposed method heightens by 28.57% at the highest, and the classification dispersion heightens by 268. 58%.