动车组监测数据分析与智能运维关键技术研究

The increasing operation costs and safety pressures are giving a serious challenge to the traditional maintenance of Electric Multiple Unit (EMU), which is mainly based on experience and theoretical model. Although the running states of EMU have been completely monitored, the equipment failures are still frequently, and on the other hand, various kinds of data-based fault diagnosis and early warning algorithms are hardly to be employed. Regarding that the data quality has become the bottleneck of data mining, this project extracts key scientific issues from the actual needs, and a scientific and technological research focusing on monitoring data will be explored. Different from traditional methods which tries to restrain the noise., a monitoring data filtering method based on deep learning and the similarity of noise is to be proposed, so that the characteristics of the system can be estimated accurately; Secondly, by mining the intermittent and heterogeneous characteristics of missing data, high density missing data recovery method based on data reliability and availability evaluation is proposed, by merging data autocorrelation and cross-correlation features. Then by breaking through the thinking pattern that data determining image, an inverse model for reconstructing data from an image is to be established. Further, in order to balance the conflicting requirements in representing the overall trend and the detailed information, a state trend analysis method based on scale transformation and iterative fusion is to be proposed. At last, a data-based EMU smart operation and maintenance system is to be developed, which can provide a scientific basis for the long-term development of China's high-speed railways.

日益增长的运维成本和安全压力，对以经验和机理模型为主的动车组运维模式提出了挑战。尽管目前已实现了动车组运行状态的全面监测，然而，一方面，设备故障依旧频发，另一方面，基于数据的故障诊断和预警算法难以落地。针对数据质量已成为数据价值挖掘的瓶颈，本项目从实际需求提炼关键科学问题，围绕监测数据分析展开科技攻关。区别于传统的将噪声“抑制”的定势思维，提取数据噪声的相似性，提出基于深度学习与相似性的动车组监测数据滤波方法，准确估计系统的属性特征；挖掘监测数据缺失的间歇性和非均匀特征，提出以数据可靠性与可用性评价为基础，以数据自相关性与互相关性特征相融合的高密度缺失数据恢复方法；突破数据决定图像的惯性思维，建立图像重构数据的逆向模型，提出基于尺度变换及迭代融合的状态趋势分析方法，有效平衡整体趋势与细节信息表征的矛盾需求，并开发基于数据的动车组智能运维系统，为保障我国高速铁路的健康长远发展提供科学依据。

本项目围绕动车组监测数据分析的核心基础，开展动车组监测数据滤波、数据恢复、自动趋势分析和动车组关键部件故障诊断与寿命预测的研究，促进人工智能在轨道交通领域的实际应用。主要研究内容包括：1）针对动车组监测数据中出现的噪声和离群值等问题，研究监测数据滤波和降维的方法,估计系统设备的运行状态；2）针对动车组缺失监测数据表现出的高密度、间歇性、非均匀特征，研究了数据与知识结合的动车组缺失监测数据恢复方法和观测缺失下列车模型的多速率异步均匀采样融合估计方法；3）分析动车组的运行特征，提出基于最优包络中心线的局部趋势提取方法和基于小波多尺度分析的局部趋势提取方法；4）针对动车组关键部件智能运维监测，研究了基于振动信号的故障诊断和考虑间歇时间的寿命预测方法等。本项目的研究成果将为保障我国高速铁路的健康长远发展提供科学依据。