面向弓网大数据的高铁在途运行安全机器学习理论与算法

With the rapid increase of high-speed railway operating mileage, the safety of high-speed railway online operating state become more prominent. Face with pantograph-catenary data that reflect high-speed railway safety increasing, how to perceive operating safety state and warn fault accurately and efficiently becomes a huge challenge for the large-scale pantograph-catenary data. Current state monitoring research is limited to superficial local single fault type recognition, and unable to obtain the potential warning of the abnormal state of high speed railway operating process. Therefore, it is urgent to solve the problem of online warn fault for the high-speed railway based on large-scale pantograph-catenary data. This project proposes a spatio-temporal modeling of the high-speed railway online operating state based on deep framework, namely the basic theory, key technologies and practical examples. First of all, to study the theory of the online operating spatio-temporal modeling of the high-speed railway for the pantograph-catenary data. Second, to study the following key technologies for online perception of the high-speed railway operating state: space structure state modeling by deep forest, time sequence state modeling by deep bi-directional long short-term memory network, and spatio-temporal cascading dynamic inference. Finally, combined with high speed railway operating state warning and prediction tasks, to verify the proposed theory and key technology. Through the above research, to solve high-speed railway online operating state modeling for the large-scale pantograph-catenary data problem , has great significance to the theoretical research and application of the rapid dynamic modeling for big data.

随着高铁运营里程的快速增长，高铁在途运行状态安全问题日益突出。面对反映高铁运营安全的弓网数据规模逐渐增大，如何在弓网大数据环境下快速对高铁在途运行状态进行有效的预警与预测是迫切需要解决的难题。本项目旨在提出一种基于深度表示的高铁在途运行状态时空模型，从基础理论、关键技术和实例验证三个方面进行研究。首先，研究面向弓网大数据的高铁在途运行安全时空模型理论框架；其次，研究空间结构状态建模的深度随机森林算法，时间序列状态建模的深度双向长短时记忆网络算法，以及空-时级联动态推理的关键技术，实现高铁在途运行状态的时空在线快速感知；最后，结合承担的高铁在途运行状态潜势预警任务，验证提出的理论与关键技术。通过上述研究，探索面向弓网大数据的高铁在途运行状态建模的难题，对大数据下快速动态时空建模理论与机器学习算法研究和应用具有重要意义。

随着高铁运营里程的快速增长，高铁在途运行状态安全问题日益突出。本项目针对反映高铁运营安全的弓网数据，主要展开四个方面的研究工作：（1）弓网数据燃弧缺陷检测与识别；（2）面向时间依赖的生成式对抗网络多标签分类研究；（3）面向全局空间关联建模的图卷积神经网络研究；（4）高铁在途信号通信传输。本项目将部分研究成果应用在高铁接触网在线检测装置中，获得较好的性能。.三年来，课题总体进展顺利，发表国际会议、期刊10篇，录取待发表2篇，申请专利1项，待申请专利2项。