基于机器学习的高铁边坡位移预测不确定性度量与应用

Uncertainty Measurement and Application of High-Speed Railway Slope Displacement Prediction Based on Machine Learning

为解决不确定性问题对高铁边坡位移预测精度的影响,引入区间预测理论量化位移预测中的不确定性问题,并建立Bootstrap-GRU-BP混合区间预测模型(BGB模型)。该模型首先采用基于Bootstrap的门控循环单元(Gated Recurrent Unit,GRU)算法度量位移预测均值和认知误差的方差,再采用BP算法度量随机误差的方差,然后将位移预测均值、认知误差和随机误差的方差3者结合在一起,量化出一定置信水平下的预测区间。最后,基于杭绍台高铁沿线边坡的监测数据,探讨BGB模型认知不确定性的响应特征,并通过对比多种区间预测模型来验证BGB模型的优越性。结果表明:BGB模型不仅能构造清晰可靠的预测区间,还能提供高精度的点预测结果;改变模型输入特征和预测算法会导致认知不确定性的改变,而BGB模型所构造的预测区间能正确地响应不确定性的变化;对比以极限学习机(Extreme Learning Machine,ELM)和支持向量回归(Support Vector Regression,SVR)为核心的区间预测模型,BGB模型的区间预测和点预测性均能更优。研究成果可为高铁边坡位移发展提供可靠的预测结果,进而为高铁边坡可靠度分析提供理论基础。

To solve the influence of uncertainty on the prediction accuracy of high-speed railway slope displacement,an interval prediction theory is introduced to quantify the uncertainty in displacement prediction,and a Bootstrap-GRU-BP(BGB)hybrid interval prediction model is established.BGB model firstly uses the Bootstrap-based Gated Recurrent Unit(GRU)algorithm to measure the predicted mean value of displacement and the variance of cognitive error,uses the BP algorithm to measure the variance of random error,and then combines the predicted mean of displacement,the variance of cognitive error and random error to quantify the prediction intervals under a certain confidence level.Finally,based on the monitoring data of the slopes along the Hangzhou-Shaoxing-Taizhou High-Speed Railway,the response characteristics of the cognitive uncertainty of the BGB model are explored,and the superiority of the BGB model is verified by comparing multiple interval prediction models.The results show that the BGB model not only constructs clear and reliable prediction intervals but also provides highly accurate point prediction results;changing model input features and prediction algorithm leads to the change of cognitive uncertainty,while the prediction intervals constructed by the BGB model can correctly respond to the changes in uncertainty.Compared to the interval prediction models centered on Extreme Learning Machine(ELM)and Support Vector Regression(SVR),the BGB model has better performance in both interval and point predictions.The research results can provide reliable prediction results for the development of high-speed railway slope displacement,and further provide the theoretical basis for the reliability analysis of high-speed railway slope.