基于GA-BP神经网络的列控车载设备故障诊断方法研究
Research on fault diagnosis method for on-board equipment of train control system based on GA-BP neural network

针对列控车载设备故障复杂且故障分析多依赖人工经验完成等问题，以车载安全计算机记录的AElog故障数据为样本，提出一种基于反向传播(Back Propagation, BP)神经网络的车载设备智能故障诊断方法。为避免初选特征信息冗余，通过主分量启发式属性约简算法，对样本进行降维降噪处理。另外，考虑到BP神经网络对初始网络权重非常敏感，以不同的权重初始化网络，往往会收敛于不同的局部极小，利用遗传算法对BP神经网络的初始权值/阈值进行优化。研究结果表明：加入属性约简提高分类器的分类性能，通过遗传算法优化的BP神经网络避免局部极小问题，且迭代步数少，降低平均误差，提高分类精度。
The on-board equipment of train control system is the core of the safe operation of high-speed rail. Due to the complexity of fault reason, fault analysis depending on the personal experience and other issues, this paper proposes an intelligent fault diagnosis method based on Back Propagation (BP), which takes AElog fault data from vehicle safety computer records as samples. In order to avoid the redundancy of features selected, the main component heuristic attribute reduction algorithm is adopted to achieve dimension reduction and noise reduction. In addition, considering that the BP neural network is very sensitive to the initial weights, thus it will converge to different local minimum when different weights are used to initialize the network. Therefore, genetic algorithm (GA) is used to optimize the initial weight and threshold of BP neural network. And finally, simulation results show that the classification performance of the classifier is improved after attribute reduction processing, and the BP neural network optimized by genetic algorithm not only avoids the local minimum problem, but also has fewer iterative steps, reduces the average error and improves the classification accuracy