基于区间二型神经模糊系统的轨道电路故障诊断  

Fault Diagnosis for Track Circuit Based on Interval Type-2 Neural-Fuzzy System

作　　者：王梓丞[1,2] 张亚东 郭进[1] 苏丽娜[1] 杨璟 宋辞 李科宏 WANG Zicheng;ZHANG Yadong;GUO Jin;SU Lina;YANG Jing;SONG Ci;LI Kehong(School of Information Science and Technology,Southwest Jiaotong University,Chengdu 611756,China;China Railway Eryuan Engineering Group Co.,Ltd.,Chengdu 610031,China)

机构地区：[1]西南交通大学信息科学与技术学院,四川成都611756 [2]中铁二院工程集团有限责任公司,四川成都610031

出　　处：《西南交通大学学报》2021年第1期190-196,共7页Journal of Southwest Jiaotong University

基　　金：四川省科技计划(2020YFG0381);中铁二院工程集团有限责任公司的科学技术研究计划资助(KYY2019040(19-21))。

摘　　要：针对目前铁路现场对轨道电路故障的判别仍然采用阈值法导致维护效率偏低的问题,提出将神经网络与模糊逻辑相结合,构建区间二型神经模糊系统(interval type-2 neural-fuzzy system,IT2NFS),通过诊断模型实现对轨道电路故障模式的智能识别.首先通过结构识别建立初步的网络结构,采用均匀设计方法生成模糊集的均值,对训练样本进行相似性测试,生成标准差和初始后件参数;再通过递归奇异值分解,优化后件参数以减小输出误差;最后,针对常见的8种故障,从实验平台上采集样本共计9000个,其中6300个样本用于模型训练,剩余2700个用于实验测试.实验结果表明:利用IT2NFS模型进行故障诊断时,每种故障类别的识别率均在82%以上,平均正确率为90.9%,仿真用时10.59 s.At present,the threshold method,despite its low efficiency,has still been used to identify the fault of track circuit on site.To handle this,an interval type-2 neural-fuzzy system(IT2NFS)was built by combining neural networks and fuzzy logic.Intelligent identification of failure modes was realized by constructing a diagnostic model.During the construction of the diagnostic model,a preliminary network structure was established through the structure identification.Uniform design method was used to generate the mean values of fuzzy sets.Then the standard deviations and initial consequent parameters were generated through performing a similarity test on training samples.At last,the optimized consequent parameters were obtained by recursive singular value decomposition to reduce the output error.For 8 common failures,a total of 9000 samples were collected from the test platform.Of them,6300 samples were used for model training,the rest 2700 samples were used for testing.The test results show that when using the IT2NFS model for fault diagnosis,the recognition rate of each fault category was above 82%,the average correct rate was 90.9%,and the simulation time was only 10.59 s.

关 键 词：轨道电路 神经网络 模糊逻辑 智能诊断 均匀设计方法 递归奇异值分解 

分 类 号：U284.2[交通运输工程—交通信息工程及控制]