基于机器学习的铁路道岔故障自动诊断方法  

Automatic fault diagnosis method of railway turnout based on machine learning

作　　者：潘亚康 PAN Yakang(Lanzhou Railway Design Institute,China Railway First Survey and Design Institute Group Co.,Ltd.,Lanzhou 730000,China)

机构地区：[1]中铁第一勘察设计院集团有限公司兰州铁道设计院,兰州730000

出　　处：《自动化与仪器仪表》2023年第2期16-20,共5页Automation & Instrumentation

基　　金：陕西省创新能力支撑计划(2019TD-014)。

摘　　要：为了提高铁路道岔故障诊断的准确性,提出一种基于机器学习的铁路道岔故障自动诊断的模型。首先采用深度森林算法作为基础算法,构建故障诊断模型。为了证明深度森林算法在道岔故障诊断方面更具优势,与SVM和BP神经网络两种经典算法模型进行对比,结果表明:深度森林算法模型的道岔故障诊断的准确率高达97%,基于SVM与BP神经网络模型的准确率分别为93%与92%;在交叉测试中深度森林算法模型的AUC面积达到0.981,在多分类测试中AUC面积为0.974。可以证明,基于深度森林算法构建的故障诊断模型诊断有较高的准确性与优越性。In order to improve the accuracy of railway turnout fault diagnosis, an automatic railway turnout fault diagnosis model based on machine learning is proposed. Firstly, the deep forest algorithm is used as the basic algorithm to construct the fault diagnosis model. In order to prove that the deep forest algorithm has more advantages in the turnout fault diagnosis, compared with the two classical algorithm models of SVM and BP neural network, the results show that the turnout fault diagnosis accuracy of the deep forest algorithm model is up to 97%, and the accuracy of the SVM and BP neural network model is 93% and 92%, respectively. In the cross-test, the AUC area of the deep forest algorithm model reached 0.981, and in the multi-classification test, the AUC area was 0.974. It can be proved that the fault diagnosis model based on deep forest algorithm has high accuracy and superiority.

关 键 词：道岔故障诊断 深度森林算法 机器学习 小数据集 

分 类 号：TP206[自动化与计算机技术—检测技术与自动化装置] U284[自动化与计算机技术—控制科学与工程]