基于LightGBM算法的沪蓉高铁列车晚点动态预测

Delay Prediction of Shanghai-Chengdu High-speed Railway Train Based on LightGBM Algorithm

高速铁路列车在运行过程中容易受到外部环境的干扰,由于干扰具备随机性、突发性和复杂性,导致列车晚点传播较快,影响范围较广。实现列车晚点精准预测对于提高调度指挥智能化水平具有重要意义。基于沪蓉高速铁路2020年全年列车运行实绩数据,运用LightGBM(Light Gradient Boosting Machine)算法,对列车运行过程中的时间相关特征、历史统计特征、前车相关特征以及其他特征等进行分析,按照影响程度提取重要特征,分别实现列车到达晚点预测和列车出发晚点预测。通过训练集数据调整算法中相关参数,结合动态预测流程实现客运列车全程到发晚点预测;使用测试集数据对模型进行验证评估。结果表明,在允许误差不超过3min的情况下,模型预测精度达到90%以上。同时与多元线性回归模型和XGBoost模型对比,结果表明LightGBM模型预测效果良好。

High-speed railway trains are easily disturbed by the external environment factors during operation.Due to the randomness,suddenness and complexity of interference,train delays spread quickly and affect a wide range.Accurate prediction of train delays is of great significance for improving the intelligence level of dispatching and commanding.Based on the actual train operation performance data of the Shanghai-Chengdu high-speed railway in 2020,the LightGBM(Light Gradient Boosting Machine)algorithm was used to analyze the time-related features,historical statistical features,and related preceding train features during the operation,and extracted important feature vector in order to build train delay prediction model.We Adjusted the relevant parameters in the algorithm through the training set data,and combined the rolling prediction process to realize the whole-process delay prediction of passenger trains.The model is verified and evaluated using the test set data.The results show that the model prediction accuracy can reach more than 90%when the allowable error does not exceed 3 minutes.At the same time,compared with the linear regression model and the XGBoost model,the proposed model has a good prediction effect.