深度时间序列异常检测在轨道状态检测中的应用研究

Deep Time Series Anomaly Detection and Its Application in Track Anomaly Detection

摘要: 轨道缺陷严重威胁高铁运行安全，传统人工巡检的方法难以实现大规模快速识别，基于振动的轨道状态监测技术因其低成本和在线监测的优势被广泛采用，针对该类时间序列数据开发高效的检测算法至关重要。本文系统综述了基于深度学习的时间序列异常检测方法及在轨道振动状态检测中的应用。首先，依据模型的学习任务类型，对基于深度学习的时间序列异常检测方法进行分类介绍，随后综述了这些方法在轨道异常检测中的具体应用，并深入探讨了时间序列异常检测在轨道异常检测应用中面临的问题与挑战，最后对未来研究工作进行展望。

Abstract: Track defects pose a serious threat to high-speed rail operational safety. Traditional manual inspection methods struggle to achieve large-scale, rapid identification. Vibration-based track condition monitoring technology, with its low cost and real-time monitoring advantages, has been widely adopted. Developing efficient detection algorithms for such time-series data is of critical importance. This paper provides a systematic review of time series anomaly detection methods based on deep learning and the application of deep learning in track vibration time series data. First, the paper categorizes and introduces time series anomaly detection methods based on deep learning according to the type of learning task of the model. Subsequently, it reviews the specific applications of these methods in track anomaly detection. The paper then delves into the challenges and issues faced in time series anomaly detection applications for track anomaly detection. Finally, it summarizes and outlooks future developments.