复杂环境下融合多源信息的高铁弓网电弧状态辨识方法研究

Arc forms when two conconductors separate and gas discharges. It stays only for a short period of time and its shape is irregular. Moreover, time when it occurs could not be predicted. Even when it occurs, it is hard to be detected. Pantograph-catenary arc is dangerous when a train is running. Besides, the cause of this problem is complicated. If the problem of arc could not be solved, the dream of raising speed of railway passenger trains would never come true. It is not easy but essential to detect the pantograph-catenary arc when the train is running even though the environment varies a lot. Nowadays, non-contact arc detection based on videos and pictures is the main trend to detect pantograph-catenary arc. If we just use a beam of light to detect the arc, it is likely that the arc might be ignored or misjudged. For example, the lightning might also be seen as the pantograph-catenary arc. This project detects the pantograph-catenary arc precisely based on the visible light, infrared ray and ultraviolet ray which is emitted when the arc occurs. First, we would divide the video into several parts which is shot when the train is running and study the way to separate the arc from the surrounding environment. Then we would analyze different kinds of fault of the pantograph-catenary and use the arc feature which might caused by two or more fault to position the pantograph-catenary arc by high precision in the future. If the research is successfully conducted, the state of the pantograph-catenary arc could be easily recognized by high pricision. In this way, we could also find the cause of pantograph-catenary arc. In the future, it might also be possible to give an alarm before the arc ossurs. All in all, it guarantees the safety and security of large-scale operations of rail transit. The research has practical and scientific significance so it is suggested to be carried out soon.

电弧是相互接触的导体分离过程中气体放电的一种形式，维持时间短、形状不规则且不可预测，检测难度大。弓网电弧对高铁可靠运行危害极大，致因复杂，是阻碍动车组提速的直接原因。在列车高速运行且环境瞬变的情况下精准辨识弓网电弧的状态是复杂的科学问题。采用基于图像的非接触检测被认为是弓网电弧辨识的必然趋势，现有研究在单一环境基于单一光源进行状态识别，会将诸如闪电等现象误判为弓网电弧且故障辨识准确率较低。本项目充分利用弓网电弧发生时所表征的可见光、红外和紫外特征，实现对弓网电弧状态的精确辨识。研究复杂环境在视频中的表征方法并实现对视频分段，研究弓网电弧故障行为并提取多个故障特征，融合多个故障特征实现对弓网电弧状态的精确辨识。项目的实施将大幅提升弓网电弧状态辨识的准确度，使得探寻弓网电弧的致因成为可能，也为弓网的“预警”研究打下基础，为弓网系统的运营提供安全保障。具有较强的紧迫性、重要的科学意义和现实意义。

高速铁路的弓网系统是高速列车获取电能的唯一方式。然而由于接触导线的不平顺、接触网的振动等多种因素，受电弓与接触导线在高速的接触与分离击穿空气产生弓网电弧。弓网电弧会给列车运行带来许多危害，例如它产生的高温会对接触导线和受电弓滑板产生严重的侵蚀和磨耗。因此，如能在高速列车复杂的运行环境下准确识别出弓网电弧，则能够及时地统计、发现弓网电弧频发区段，指导高速列车的运维工作，为我国铁路弓网长期全生命周期管理提供可靠保障。. 复杂多变的高速列车运行环境给非接触式弓网电弧辨识带来极大的挑战，项目创造性地提出基于弓网视频实现高速列车运行环境的感知，进而优化弓网电弧辨识方法。本项目将以卷积神经网络为代表的人工智能技术应用于高速列车的运行环境辨识中，建立了高速列车运行环境的数据集，并据此建立了高精度的高速列车运行环境辨识模型。本项目解析了弓网电弧在可见光视频和红外视频数据中不同的呈现规律，实现了不同数据源下的弓网电弧辨识。项目建立了考虑列车运行环境下的融合多源信息的弓网电弧辨识模型。根据列车运行环境，动态更新多源信息的可靠性权重，并基于加权证据理论对可见光和红外的检测结果进行决策级信息融合。. 经复兴号的弓网视频数据验证，环境辨识模型的准确度达到98%；基于可见光的弓网电弧识别准确度为89%；信息融合后的弓网电弧识别准确度97.5%。围绕该项目，发表了SCI/SSCI检索国际期刊论文10篇，申请发明专利3项，培养研究生7名。项目的研究成果为弓网电弧的辨识提供了切实可行的解决方案，项目的研究方法在轨道交通信息感知领域得到了推广应用。项目可见光特征提取部分采用的基于深度学习的视觉信息提取方法，成功应用于道岔故障诊断、缺口识别和司机行为识别，并发表了高水平成果；项目基于可见光和红外的融合方法，成功应用于道路交叉口的全天候行人识别。项目的研究对于复杂环境下基于视觉的状态感知有重要的理论意义，对于轨道交通智能运维有重要的现实意义。