摘要
轨道电路作为铁路信号的重要设备之一,易受环境、电气干扰等因素影响而发生故障,一旦发生故障将直接影响列车运行的安全性。目前,传统的轨道电路故障诊断方法在提取故障数据的全局和局部特征上仍存在局限性,而人工智能与深度学习方法的兴起为解决这一问题提供了新的解决思路。鉴于此,以ZPW-2000A轨道电路为研究对象,提出一种基于CNN,LSTM和注意力机制的故障诊断方法,记作CNN-LSTM-Attention。具体来说,该方法通过CNN提取轨道电路故障的局部特征;通过LSTM挖掘时间序列数据的相关性和时间依赖关系,进一步获取全局特征;再引入注意力机制对特征赋予不同的权重,最终实现轨道电路故障的智能诊断。最后,依托信号集中监测系统,获取ZPW-2000A型轨道电路的31种典型故障模式,模拟故障曲线并生成前、中、后3个区段的时间序列数据集。在该数据集上进行实验验证。结果显示:与传统的时序数据深度模型,如Multi-LSTM、CNN-LSTM相比,CNN-LSTM-Attention可以提取时序数据的关键特征和全局特征,在测试集上的诊断性能是最佳的,准确率达到99.9%以上。而且注意力权重热力图显示模型更关注输入的中心部分,这说明本区段数据是影响故障发生的关键因素。CNN-LSTM-Attention为轨道电路故障诊断和铁路系统的安全运行提供了重要的理论和技术支撑,有一定的应用和推广价值。
Track circuit is one of the important railways signaling equipment,which will directly affect the safety of train operation in case of failure.Traditional fault diagnosis methods have limitations in extracting global and local features of faults,and the rise of deep learning methods provides new solutions to solve this problem.This paper proposed a fault diagnosis method based on CNN,LSTM and Attention mechanism,denoted as CNNLSTM-Attention,with ZPW-2000A as the research object.Specifically,the method extracted local features of track circuit faults by CNN,mined correlation and global features of time series data by LSTM,and then introduced Attention mechanism to assign different weights to the features,and finally realized fault diagnosis.Finally,this paper collected 31 common faults of ZPW-2000A rail circuit,simulates the fault curves and generates a data set.Experimental validation was carried out on this data set.The results show that compared with mainstream methods,CNN-LSTM-Attention has the best diagnostic performance and strong generalization ability.In conclusion,the method proposed in the paper can provide important theoretical and technical support for track circuit fault diagnosis and safe operation of railroad systems.
作者
杨勇
可婷
胡启正
张志敏
YANG Yong;KE Ting;HU Qizheng;ZHANG Zhimin(Signal and Communication Research Institute,China Academy of Railway Sciences Corporation Limited,Beijing 100081,China;College of Artificial Intelligence,Tianjin University of Science&Technology,Tianjin 300457,China;Research Center for Intelligent Operation and Maintenance of Communication Signal Infrastructure in Railway Industry Engineering,Beijing 100081,China)
出处
《铁道科学与工程学报》
北大核心
2025年第5期2380-2392,共13页
Journal of Railway Science and Engineering
基金
中国国家铁路集团有限公司科研项目(L2022G004)。
