基于影响因子聚类与分布拟合的补偿电容剩余寿命预测

Residual life prediction of compensation capacitors based on cluster of influence factors and distribution fitting

补偿电容是保证无绝缘轨道电路正常传输的重要设备,广泛安装于我国高速线路。为了研究补偿电容在实际运用环境下的剩余寿命,提出一种影响因子聚类与分布拟合相结合的补偿电容剩余寿命预测方法。首先以补偿电容维修数据为研究对象,建立补偿电容退化趋势数据模型;分析补偿电容功能退化潜在影响因子,以这些影响因子作为特征属性建立数学模型描述,基于补偿电容历史数据构建特征样本集;通过聚类算法对特征样本集进行类簇划分得到多个子类特征样本集,通过极大似然估计法针对每个子类特征样本集计算威布尔分布拟合参数;通过补偿电容与各子类特征样本集的隶属度以及计算的分布曲线计算补偿电容剩余寿命。为了验证本文所提方法的性能,分别与采用LSTM的补偿电容故障预测方法、威布尔分布拟合的方法在单个补偿电容剩余寿命预测及补偿电容故障数量预测方面进行准确性对比试验。基于27 000余条补偿电容实际维修数据的实验结果表明:1)在单个补偿电容剩余寿命预测方面,所提方法与威布尔分布方法的预测准确性最高分别为95.0%、87.0%,且随着测试样本数量的增加,本文所提方法性能降低速度更慢;2)在补偿电容故障数量预测方面,所提方法的正确率相比基于LSTM和威布尔分布拟合的故障数量预测方法分别提高5.6个百分点和41.3个百分点。

Compensation capacitor is an important equipment to ensure the normal transmission of jointless track circuit,which is widely installed in high-speed lines in China.In order to study the residual life of a compensation capacitor in the actual environment,a method based on cluster of influence factors and distribution fitting was proposed.Firstly,a data model of degradation trend of compensation capacitors was established based on the maintenance data.The potential influence factors related to the function degradation of compensation capacitors were analyzed.A mathematical model was established by taking these influence factors as characteristic attributes.The characteristic sample set was constructed based on historical maintenance data.Multiple characteristic sample sets were clustered by clustering algorithm to obtain sub-class characteristic sample sets.For each sub-class characteristic sample set,the fitting parameters of Weibull distribution were carried out by maximum likelihood estimation method.Finally,the residual life was calculated by the membership degree of a compensation capacitor and each sub-class characteristic sample set and the calculated distribution curve.In order to verify the performance of the method proposed in this paper,the accuracy of the residual life prediction of a single compensation capacitor and the fault number prediction of the of compensation capacitors are compared with methods using LSTM and Weibull distribution,respectively.The experimental results based on more than 27000 pieces of actual maintenance data of compensation capacitors are shown as follows.(1)In the prediction of the residual life of a single compensation capacitor,the highest prediction accuracy of the proposed method and the Weibull distribution method is 95.0%and 87.0%respectively,and the performance of the proposed method decreases more slowly with the increase of test samples.(2)In the fault number prediction of compensation capacitors,the accuracy of the proposed method is improved by 5.6 percentage points and 41.3 percentage points respectively compared with the fault number prediction methods based on LSTM and Weibull distribution.