基于CHMM(连续隐马尔科夫模型)的转辙机故障预测方法

Research on Switch Machine Fault Prediction Based on CHMM

[目的]转辙机作为城市轨道交通信号设备的重要组成部分,一旦发生故障,会对运营造成重要影响,对其健康状态的监测和预测显得尤为重要。[方法]提出了一种基于CHMM(连续隐马尔科夫模型)的转辙机故障预测方法。提取转辙机退化状态特征,并基于t-SNE算法对原始输入数据进行降维,减少余特征;利用谱聚类算法确定最优退化状态数目,进行聚类分割,分析转辙机动作功率曲线退化状态特征;基于CHMM模型并与故障诊断模型相结合,构建退化状态识别模型与故障识别模型,实现对转辙机的故障预测。以实测数据作为研究对象,对基于CHMM的转机故障预测方法进行试验验证。[结果及结论]该方法以转辙机正常动作功率曲线为研究对象,深入挖掘监测数据,提取的退化状态特征具有良好的表征能力。根据严重退化状态下曲线模型与正常曲线模型匹配结果,在转辙机功率发生异常波动时,可实现对转辙机故障类型的预测。

[Objective]As an important component of urban rail transit signal equipment,once the switch machine malfunc-tions,the operation will be seriously affected.Monitoring and predicting its health status is particularly important.[Method]A fault prediction method for switch machine based on CHMM(continuous hidden Markov model)is proposed.The features of the switch machine degradation state are extracted,and the original input data dimension is reduced based on t-SNE algo-rithm to reduce the redundant features.Spectral clustering algo-rithm is used to determine the optimal number of degradation states,make clustering segmentation and analyze the degrada-tion state features of the switch machine action power curve.Based on CHMM model and fault diagnosis model,the switch machine fault prediction is realized by constructing degradation state identification model and fault identification model.The fault prediction method for switch machine based on CHMM is verified through measured data.[Result&Conclusion]With the normal operation power curve of the switch machine as the research object,the above method taps the monitored data deeply,and the extracted degradation state features have good expressive ability.According to the matching results between the curve model in severely degraded state and the normal curve model,the fault types of switch machine can be predic-ted when the power of the switch machine fluctuates abnormal-y.