基于DBN-MPA-LSSVM的无绝缘轨道电路故障诊断研究

针对区间无绝缘轨道电路故障类型复杂、诊断精度低等问题,从故障特征提取和特征分类两方面出发,提出了一种深度置信网络(DBN)和海洋捕食者算法(MPA)优化最小二乘支持向量机(LSSVM)的故障诊断方法。 首先,将集中监测数据和状态标签输入到 DBN,以半监督的方式进行降维和特征提取,从而挖掘轨道电路不同故障特征信息;然后,采用 MPA 智能算法对LSSVM 的惩罚因子和核函数参数进行寻优并建立最优 MPA-LSSVM 诊断模型;最后,将 DBN 提取的特征样本导入诊断模型进行轨道电路的故障分类识别。 DBN-MPA-LSSVM 诊断模型充分利用了 DBN 在特征提取过程中的逐层提取优势以及 LSSVM 在解决小样本情况下高维模式识别的优势。 实验验证与对比分析表明,DBN-MPA-LSSVM 模型测试集准确率为 98. 33%,MPA 优化算法较 PSO、GWO、GA 算法模型诊断准确率分别提高了 6. 11%、3. 89%、3. 33%,平均准确率为 97. 98%,为基于数据驱动的轨道电路故障诊断技术提供了一种新的方法。

Research on fault diagnosis of jointless track circuitbased on DBN-MPA-LSSVM

Aiming at the problems of complex fault types and low diagnosis accuracy of section jointless track circuit, a fault diagnosismethod of least squares support vector machine(LSSVM)optimized by deep belief network(DBN)and marine predators algorithm (MPA)is proposed from the two aspects of fault feature extraction and feature classification. Firstly, the centralized monitoring data and statuslabels are input into DBN, and the dimensionality reduction feature extraction is carried out in a semi supervised way, so as to mine thedifferent fault feature information of track circuit. Then, the intelligent algorithm MPA is used to optimize the penalty factor and kernelfunction parameters of LSSVM, and the optimal MPA-LSSVM diagnosis model is established. Finally, the feature samples extracted byDBN are introduced into the diagnosis model for fault classification and identification of track circuit. DBN-MPA-LSSVM diagnosticmodel makes full use of the advantages of layer by layer extraction of DBN in the process of feature extraction and the advantages ofLSSVM in solving high-dimensional pattern recognition in the case of small samples. Experimental validation and comparative analysisshow that the DBN-MPA-LSSVM model test set accuracy is 98. 33%, and the MPA optimization algorithm improves the diagnosisaccuracy by 6. 11%, 3. 89%, and 3. 33% compared with PSO, GWO, and GA algorithm models, respectively, with an averageaccuracy of 97. 98%, which provides a new data-driven rail circuit fault diagnosis technology based on method.