摘要
为了合理预测超高路基沉降并推断其上部路面水稳层的铺筑时间,结合ANN算法的非线性映射能力和GA算法的全局优化搜索能力,针对湖南省某山区高速公路高填路基的监测数据建立了机器学习GA-ANN模型,预测了后续路基沉降曲线,对比了实测数据、ANN预测数据、GA-ANN预测数据,验证了采用GA-ANN进行路基沉降预测的优势,并根据预测沉降曲线推断水稳层铺筑时间。结果表明,GA算法可以为ANN提供更合理的权重阈值。相对ANN,GA-ANN模型的RMSE和MAPE更小,预测精度更加理想,它更好地反映了沉降速率的突变,而ANN容易造成对沉降提前收敛的乐观估计;利用机器学习预测得出的沉降曲线,可以作为路面水稳层铺筑时间初步确定的推断依据,有助于施工单位提前制定施工计划和做好施工准备;水稳层铺设过程中未出现开裂现象,该项目已经通车半年,路面运营状态良好,为类似高填路基甚至超高填路基的水稳层铺筑提供了借鉴参考。
In order to reasonably predict the settlement of ultra-high subgrade and deduce the construction time of water-stable layer above,this paper combines the nonlinear mapping ability of the ANN algorithm with the global optimization search capability of the GA algorithm,a machine learning GA-ANN model was established based on the monitoring data of a high-filledsubgrade in a mountainous area of Hunan Province,the subsequent settlement curve of the embankment was predicted,and the measured data,ANN predicted data,and GA-ANN predicted data were compared,the advantages of using GA-ANN for embankment settlement prediction were verified,and according to the predicted settlement curve,the timing of the water-bound macadam layer construction was inferred.The results show that the GA algorithm can provide more reasonable weight bias values for the ANN.Compared to the ANN alone,the GA-ANN model has smaller RMSE and MAPE,and greater prediction accuracy,it better reflects the sudden changes in the settlement rate,while the ANN tends to optimistically estimate early convergence of settlement;the settlement curve predicted using machine learning can serve as a preliminary basis for inferring the construction time of water-stable layer,which helps construction units to formulate construction plans and prepare for construction in advance;no cracking occurred in the water-stable layer during the paving process.The project has been open to traffic for half a year,and the operation status of the road surface is good,thereby providing a reference for the construction of water-bound layers in similar high and even ultra-high fill embankments.
作者
熊辉
聂伟
XIONG Hui;NIE Wei(Hunan Railway Construction Investment Co.,Ltd.,Changsha,Hunan 410017,China;Hunan Communications Research Institute Co.,Ltd.,Changsha,Hunan 410114,China)
出处
《公路工程》
2025年第3期178-185,共8页
Highway Engineering
基金
湖南省交通科技项目(202214)。
关键词
路基
沉降预测
机器学习
人工神经网络
遗传算法
水稳层
subgrade
settlement prediction
machine learning
artificial neural network
genetic algorithm
water-stable layer
