基于深度置信网络的轨道电路剩余寿命预测

Residual life prediction of track circuit based on depth believe network

通过对影响轨道电路运行状态的设备进行分析,构建深度置信网络对微机监测系统记录的轨道电路运行数据及实测数据进行特征提取。采用权重分配的方式结合多种影响设备运行状态的因素计算反映轨道电路运行状态的健康评估指标,通过健康评估指标对轨道电路的运行状态进行划分,根据全生命周期的历史运行数据构建其各个状态的隐半马尔可夫模型。结合深度置信网络对轨道电路的退化状态和剩余寿命进行仿真试验。研究结果表明:采用融合深度置信网络进行特征提取后训练的隐半马尔可夫模型进行剩余寿命预测准确度和退化状态识别率相比原始隐半马尔可夫模型有较大的提高。

In this paper,through the analysis of the equipment that affects the running state of the track circuit,the in-depth confidence network was constructed to extract the characteristics of the running data and the measured data of the track circuit recorded by the microcomputer monitoring system,and the health assessment index reflecting the running state of the track circuit was calculated by the way of weight distribution combined with a variety of factors that affect the running state of the equipment,and the track was evaluated by the health assessment index.The operation state of the circuit was divided.The hidden semi-Markov model of each state was constructed according to the historical operation data of the whole life cycle,and the degradation state and the remaining life of the track circuit were simulated with the depth confidence network.The test results show that the hidden semi-Markov model trained after feature extraction is used to predict the remaining life.Compared with the original hidden semi Markov model,the accuracy and the degraded state recognition rate are greatly improved.