铁道科学与工程学报2025，Vol.22Issue(7)：2921-2931,11.DOI:10.19713/j.cnki.43-1423/u.T20241549
基于内外双视角的高速铁路风险预测
High-speed railway risk prediction based on internal and external perspectives
摘要
Abstract
Accurate prediction of high-speed railway risks is crucial for the safety management of such systems.To effectively forecast the probability of risks during high-speed railway operations and address the challenge of simultaneously capturing both internal and external characteristics of accident causes,an Internal and External Perspectives on the Topological Dendrogram of Accident Causes(IEPTDAC)was proposed for high-speed railways.Firstly,the topological relationships of internal accident causes were characterized based on a tree-like structure,and the external characteristics of accident causes were extracted from the aspects of human,machine,environment,and management.Then,a multi-layer convolution operation of Convolutional Neural Network(CNN)was employed to extract both internal and external features of accident causes.Additionally,the Particle Swarm Optimization(PSO)algorithm was introduced to optimize the key hyperparameters of the CNN,further enhancing the model's predictive performance.Finally,five sections were selected from a railway administration group company,taking data on 19 accident causes and risk accidents as the research objects.Under the time granularities of 1 hour,3 hours,and 5 hours,a comparative analysis was conducted on the IEPTDAC model and the 9 existing prediction models respectively.The experimental results demonstrated that the IEPTDAC model exhibited superior prediction accuracy and fitting performance compared with both existing combination prediction models and traditional single prediction models.For instance,at a 1-hour time granularity,when compared to the prediction model based on Transient Extraction Transform and DSRNet-AttBiLSTM in the control experiment,the IEPTDAC model achieves a reduction in Mean Absolute Error(fmae)by 32.04%,a decrease in Root Mean Square Error(frmse)by 36.35%,and an increase in the Coefficient of Determination(fr2)by 0.46%.Across the 1-hour,3-hour,and 5-hour time granularities,the IEPTDAC model improves the fr2 by 1.71%,3.00%,and 1.27%,respectively,compared to the traditional Convolutional Long Short-Term Memory(ConvLSTM)model.Furthermore,ablation experiments designed in this paper validate the rationality and effectiveness of each branch of the IEPTDAC model.This method can provide an effective technical means for high-speed railway risk prediction.关键词
高速铁路/卷积神经网络/深度学习/多尺度风险预测/粒子群优化算法Key words
speed railway/convolutional neural networks/deep learning/multi-scale risk prediction/particle swarm optimization algorithm分类
交通工程引用本文复制引用
夏溪蔓,孟学雷,程晓卿,林立,韩正..基于内外双视角的高速铁路风险预测[J].铁道科学与工程学报,2025,22(7):2921-2931,11.基金项目
甘肃省科技计划资助项目(24JRRA865) （24JRRA865）
国家自然科学基金资助项目(72361020) （72361020）
