基于SOM-BP混合神经网络的道岔设备退化状态研究

针对道岔设备故障频繁、维修成本高,且现有基于故障数据的诊断方法无法描述道岔退化过程,难以实现故障超前预判的问题,进行基于SOM-BP混合神经网络的道岔设备退化状态研究。依据采集的道岔非故障功率数据按区段提取峰值、方差、峭度等特征参数,基于平均影响值进行特征参数选择,并确定输入维数;使用自组织特征映射(SOM)神经网络对输入特征参数进行多次聚类学习,分析学习结果得到6种退化状态样本数据;构建15-13-6型BP神经网络结构模型,实现对道岔设备退化状态的识别。结果表明,采用SOM-BP混合神经网络进行道岔设备退化状态识别的准确率达到95.56%。

Research on Degradation State of Turnout Equipment Based on SOM-BP Hybrid Neural Network

In view of the frequent faults and high maintenance costs of turnout equipment,the existing diagnosis methods based on fault data can not describe the degradation process of turnout,and it is difficult to achieve the fault prediction in advance,the degradation state of turnout equipment based on SOM-BP hybrid neural network is studied.According to the collected non fault power data,the peak value,variance,kurtosis and other characteristic parameters are extracted according to the section,and the characteristic parameters are selected based on the mean influence value(MIV),and the input dimension is determined.Self-Organizing feature Map(SOM)neural network is used to cluster the input feature parameters for many times,and six degraded state sample data are obtained by analyzing the learning results.The structure model of 15-13-6 BP neural network is constructed to recognize the degradation state of turnout equipment.Results show that the recognition accuracy of SOM-BP hybrid neural network to identify the degradation state of turnout equipment is 95.56%.