基于数据的机车牵引变流器故障预测及安全对策研究

As one complex nonlinear system, fault prediction and safety precautions of locomotive power converter are directly related to safe operation of the whole high-speed railway. This project investigates signal processing technology, fault prediction model and safety precautions. Based on historical and current data, as well as expert experiences, a data-based fault prediction theory and safety precautions of locomotive power converter are studied by using the approaches of fault prediction technique, wavelet transform, support vector machine and Markov decision. The research process can be divided into three steps: I. The state characteristic signals are extracted by using wavelet transform to set up the independent feature vector sets, on the basis of research on fault evolution mechanism of components, II. A data-based support vector machine prediction model is established in consideration of multi-class, non-linearly classification and non-uniform distribution characteristics of sampled data; III. Safety precautions including the corresponding condition-based maintenance and fault-tolerant control schemes are presented according to the moderative and aggravated degradations of components. The research of this project has great significance for improving the safety and reliability operation of high-speed railway transportation, on the occasion of other requirements for reliable operation of power electronics, the research results still have important reference value and application prospect.

机车牵引变流器作为一个复杂非线性系统，其各部件故障的提前预警以及安全对策直接关系到整个高速铁路的行车安全。本项目以信号处理技术、故障预测模型、安全对策为研究内容，基于系统的历史数据、当前数据和专家经验信息等，以故障预测技术为途径，以小波变换、支持向量机、马尔可夫决策等理论和技术为手段，建立基于数据的机车牵引变流器故障预测及安全对策方法。研究过程可分为三步：1.在研究部件故障演化机理的基础上，以小波变换为手段对状态特征信号进行提取，建立具有独立性的特征向量集；2.面对样本数据多类别、线性不可分、不均匀分布等特点，建立基于数据的支持向量机预测模型；3.针对部件缓慢劣化和加剧劣化的情形，分别给出相应的视情维修和容错控制方案等安全对策。本项目的开展对提高我国高速轨道交通的安全性与可靠性具有十分重大的意义，同时对其它要求电力电子可靠运行的场合，也具有重要的参考价值和应用前景。

机车牵引变流器作为一个复杂非线性系统，其各部件故障的提前预警以及安全对策直接关系到整个高速铁路的行车安全。本项目重点围绕核心部件故障预警、检测、识别，以及维修决策、容错控制等两个领域开展研究工作以提高系统可靠性。针对前者，项目组分别提出了基于数据驱动技术，以“信号处理+特征量提取及优化组合+模式分类”为思路的功率器件故障诊断方案；基于电路模型分析理论，以具有电气含义的特征量为诊断与识别依据的功率器件故障诊断方案；基于拓扑结构对称性，以相似性距离函数值为特征量的功率器件故障诊断方案；以及基于观测器的传感器故障预警、诊断、识别与系统重构的研究方案，取得了一系列富有创新性的研究成果，可以实现故障快速、精确、强鲁棒性的诊断。针对后者，项目组建立了多目标多约束的维修决策和资源优化调度模型；重点研究了多种非冗余容错拓扑功率变换器的建模与控制方法，提出了一系列高性能容错策略，取得了系列突出研究成果，成功实现了故障情况下系统持续高效安全运行，解决了故障前后系统实时平滑切换和重构问题。本项目的研究成果可以实现诸如机车牵引变流器这类典型功率变换系统故障的在线检测、识别、隔离与容错，对提高这类系统的可靠性与安全性具有重要的工程应用价值。