自适应粒子群优化的HMM故障诊断方法及应用

A fault diagnosis method with application of HMM Based onadaptive particle swarm optimization

针对传统隐马尔科夫模型(hidden Markov model,HMM)解决学习训练过程中参数容易局部收敛的问题,采用一种自适应粒子群算法对HMM模型进行优化改进。在基础粒子群算法中加入惯性权重因子,并根据算法迭代结果对算法各因子大小进行动态控制,增强该算法的全局搜索能力。改进后的粒子群算法融入HMM模型训练过程参数学习的优化过程,能够提高HMM的训练精度。将其应用到油机电站的故障诊断当中,通过对其实测振动数据进行分析,与标准算法相比,对油机处于正常、供气不足、进气门间隙异常状态的分类准确率都有所提升,整体诊断精度达到97.3%。结果表明,基于自适应粒子群优化的HMM故障诊断方法能够有效解决传统模型的局部收敛问题。

Aiming at the problem that parameters learning algorithm of hidden Markov model(HMM)easily converges to local optimal solutions,an improved HMM was proposed based on adaptive particle swarm optimization(PSO).An inertial weight factor was joined in the basic PSO.It was controlled dynamically based on the iterative results of the algorithm.Global search capability of the new algorithm thus was improved.The adaptive method was integrated to parameters learning algorithm of HMM,which contributed to optimize the initial parameters of HMM.The new model was applied to fault diagnosis for diesel engine with normal state,lack of air supply and intake valve clearance fault.Actual engine vibration data were analyzed.The correct classification rate of the proposed method reaches 97.3%,better than the traditional method based on PSO.The results verify that the proposed fault diagnosis method has a better performance than the traditional algorithm.