基于隐马尔科夫模型的牵引变压器油纸绝缘剩余寿命预测

Residual life prediction of oil-paper insulation for traction transformersbased on hidden Markov model

为考虑动车组车载牵引变压器油纸绝缘老化状态对剩余寿命的影响,本文采用油中溶解气体含量作为反映油纸绝缘老化的特征参量,由于油纸绝缘老化过程中具有动态特性,依据油纸绝缘老化状态与特征参量之间对应的隐藏状态与观测状态建立隐马尔科夫模型。首先通过贝叶斯信息准则确定隐藏状态数目,利用k-Means算法对观测序列进行聚类处理。在此基础上,利用Viterbi算法对观测序列进行状态识别进而确定油纸绝缘老化状态,最后结合比例风险模型确定可靠度函数。通过CRH2型动车组车载牵引变压器油中溶解气体在线监测数据验证模型的准确性,最终得到油纸绝缘的剩余寿命与实际寿命基本相符,从而可为后续制定牵引变压器维修策略提供理论参考。

In order to study the influence of ageing status of oil-paper insulation of EMU on-board traction transformer on its residual life,the dissolved gas content in oil was used as the characteristic parameter to reflect the ageing of oil-paper insulation in this paper.Since the ageing process of oil-paper insulation has dynamic characteristic,a hidden Markov model was established according to the corresponding hidden state and observation state between the ageing status of oil-paper insulation and the characteristic parameters.Firstly,the number of hidden states is determined using Bayesian information criteria,and then the observation sequence is clustered by using the k-Means algorithm.On this basis,the state of the observation sequence is identified by the Viterbi algorithm to determine the ageing state of oil-paper insulation.Finally,the reliability function is determined by combining the proportional risk model.The accuracy of the model is verified according to the online monitoring data of the dissolved gases in the oil of on-board traction transformer of CRH2 EMU,and it is obtained that the remaining life of the oil-paper insulation is basically consistent with the actual life,which can provide theoretical reference for the subsequent development of traction transformer maintenance strategy.