高亚丽 陈光武

摘  要：道岔是直接关系到高速铁路安全运营的关键信号设备，目前对其控制电路故障诊断方式主要是依靠简单仪器与人的以往经验。针对道岔控制电路故障的模糊性与不确定性特点和实现道岔控制电路故障诊断的高准确率与智能化，文章提出一种基于模糊推理法与模糊神经网络（Fuzzy Neural Network，FNN）相结合的诊断方法：首先在总结S700K道岔控制电路常见的8种故障类型、建立模糊规则和选择隶属函数的基础上，建立基于模糊推理法FNN诊断模型;其次为避免算法收敛慢和陷入局部最优，文章通过引入改进的动态学习速率和附加动量项的學习调整法，实现对算法的改进，并比较算法改进前后的误差曲线;最后将文中算法和BP神经网络进行诊断准确率对比。仿真实验结果表明，该算法对道岔控制电路故障诊断的准确率高并且能达到99.6%以上。

关键词：道岔控制电路;故障诊断;模糊推理法;模糊神经网络

中图分类号：U284.92       文献标志码：A              文章编号：2095-2945（2020）15-0125-03

Abstract： Turnout is a key signal device directly related to the safe operation of high-speed railways. At present， the fault diagnosis method of its control circuit mainly relies on the past experience of simple instruments and people. Aiming at the ambiguity and uncertainty of turnout control circuit faults and the high accuracy and intelligence of turnout control circuit fault diagnosis， this paper proposes diagnosis method combiningFuzzy Inference Method and Fuzzy Neural Network （FNN）： Firstly， based on the summary of the eight common fault types of S700K turnout control circuit， the establishment of fuzzy rules and the selection of membership functions， a FNN diagnosis model based on fuzzy inference is established; secondly， in order to avoid the algorithm's slow convergence and falling into local optimum， this paper introduces an improved dynamic learning rate and a learning adjustment method with additional momentum terms， then the algorithm is improved， and the error curves before and after the algorithm improvement are compared. Finally， the diagnostic accuracy of the algorithm in this paper is compared with the BP neural network. Simulation results show that the algorithm has high accuracy for the turnout control circuit fault diagnosis and can reach more than 99.6%.

Keywords： switch control circuit; fault diagnosis; Fuzzy Inference Method; Fuzzy Neural Network （FNN）

现如今，对于铁路关键信号设备之一的转辙机，如何对其高效、准确的故障诊断已经成为一个刻不容缓的课题。目前现场其故障诊断主要依靠工作人员观察微机监测系统所采集的转辙机动作电流或功率曲线来实现故障识别，这样的方法既不能智能化的分析出其故障类型，也不能智能化分析出其故障原因。

近年来，针对道岔故障诊断，国内外专家和学者提出不同方法，但都有不足。文献[1]提出灰关联道岔故障诊断方法，该方法无需训练即可对道岔故障进行诊断，不足之处是所建模型不具备自组织、自适应能力。……