基于机器学习的轨道电路状态判断

初广前，李璐*，张嘉驰，钱佳瑶

山东交通学院轨道交通学院，山东 济南  250357

摘要：为准确判断轨道电路的状态，保障轨道交通的安全运行，研究无监督学习中的高斯混合模型（Gaussian mixed model，GMM）、有监督学习中的反向传播神经网络（back propagation neural network，BPNN）模型与逻辑回归（logistic regression, LR）分类模型等3种典型机器学习算法在轨道电路状态判断中的应用。将轨道电路分为占用状态和空闲状态，应用3种算法实现状态分类任务，并在实际测得的数据集上对比3种算法的分类性能。测试结果表明：3种算法均能实现轨道电路状态的准确分类。与BPNN模型和LR分类模型相比，GMM无需训练过程，可降低人工成本。

关键词：轨道电路；状态分类；GMM；BPNN；LR分类模型

State judgement for track circuits based on machine learning

CHU Guangqian, LI Lu*, ZHANG Jiachi, QIAN Jiayao

School of Rail Transportation, Shandong Jiaotong University, Jinan 250357, China

Abstract:In order to precisely judge the state of track circuit and guarantee the safe operation of rail traffic, the utilization of three typical machine learning algorithms, such as the gaussian mixed model (GMM) of unsupervised learning, the back propagation neural network (BPNN) and the logistic regression (LR) classification model of supervised learning are studied. The track circuits are divided into the occupancy state and idle state. GMM, BPNN, and LR are applied to accomplish the state classification. The classification performance of three machine learning algorithms is compared based on practical data. The results show that the three algorithms can achieve accurate classification of track circuit states. Compared with BPNN and LR, GMM require no training process, which can greatly reduce labor costs.

Keywords:track circuit; state classification; GMM;BPNN; LR classification model