集成学习在运营隧道病害预测中的研究与应用

Research and Application of Ensemble Learning in the Prediction of Operational Tunnel Diseases

轨道交通运营安全是城市安全的重要保障。隧道是城市轨道交通的主要形式之一,其结构安全受地质条件、施工质量等各种复杂因素影响,各种因素之间会形成复杂的非线性关系。及时的预警隧道结构病害,有利于降低地铁运营风险,保证地铁隧道长期运营。传统的隧道结构安全评估是从发现已有的病害出发,评估分析已有病害发展规律,再提出维护方案,目前还末有直接预测病害发生的相关研究。机器学习算法可以综合考虑各种影响因素,通过训练模型逼近病害与其影响因素之间的复杂非线性关系,从而预测下一阶段病害的发生。利用机器学习算法中的两种典型的集成学习算法-随机森林算法和XGBoost算法,训练隧道结构病害预测模型,并通过预测结果与真实结果的对比,根据准确率、召回率、F1值等评价指标,评估预测模型效果。实验证明,预测模型准确率可达90%以上,F1值可达80%以上,预警模型可为政府和隧道维保部门提供辅助决策支持。

The operation safety of rail transit is an important guarantee for the safety of the city.Tunnel is one of the main forms of urban rail transit.Its structural safety is affected by various complex factors,such as geological conditions and construction quality,etc.,among which a complex nonlinear relationship is formed.Timely warning of tunnel structural diseases is conducive to reducing the risk of subway operation and ensuring the long-term operation of subway tunnels.Traditional tunnel structure safety assessment starts from the discovery of existing diseases,evaluates and analyzes the development tendency of existing diseases,and then proposes the maintenance plan.According to the present papers,it is a pity that no relevant research has been found to directly predict the occurrence of diseases so far.From another aspect,the machine learning algorithm can comprehensively consider various influencing factors and approach the complex nonlinear relationships between the disease and its influencing factors through the training model,so as to predict the diseases of the next stage.Two typical ensemble learning algorithms in machine learning-random forest algorithm and XGBoost algorithm are used to train the tunnel structure disease prediction model.By comparing the predicted results with the real results,the prediction models are evaluated by the evaluation indexes such as accuracy,recall rate and F1 value.The experimental results show that the accuracy of the prediction model can reach more than 90%,and the F1 value can reach more than 80%.The early warning model can provide auxiliary decision support for the government and tunnel maintenance department.