基于KNE-BPNN的电务设备故障预测

Fault prediction for communication and signal equipment based on KNE-BPNN

针对铁路电务设备故障频发、运行效率低且无有效故障预测方法等现实问题,提出一种基于K-均值-邻域近似条件熵与BP神经网络(KNE-BPNN)的电务设备故障预测模型。首先,采用基于K-均值聚类的样例约简算法约简设备故障决策表中的冗余样例;其次,运用邻域近似条件熵属性约简方法对样例约简后故障决策表中的非必要属性进行约简;最后,使用经过样例和属性约简后的样本集训练BP神经网络并进行模型预测,直到模型输出结果满足预设条件为止。实验结果表明KNE-BPNN故障预测模型的预测精度和泛化性能均满足电务设备管理的实际需求。

Based on the practical problems,such as frequent failures,low operating efficiency and lacking of effective fault prediction methods for railway communication and signal( C&S) equipment,this paper proposed a fault prediction model for C&S equipment based on K-means-neighborhood approximate conditional entropy and BP neural network( KNE-BPNN).Firstly,it used a sample reduction algorithm based on K-means clustering to reduce the redundant samples in the equipment failure decision table. Secondly,it used neighborhood approximate conditional entropy attribute reduction theory to reduce the non-essential attributes in the fault decision table after sample reduction. Finally,it trained the BP neural network by using the reduced sample set,and trained the model until its output met the expected requirement. The experimental results show that the prediction precision and generalization performance of the KNE-BPNN fault prediction model can meet the actual requirements.