基于支持向量机与神经网络法的路基沉降预测对比研究  

Comparative Study on Prediction of Subgrade Settlement Based on Support Vector Machine and Neural Network Methods

作　　者：赵志刚 左仕 李清 

机构地区：[1]中铁七局集团有限公司,郑州450016 [2]中南大学土木工程学院,长沙410075

出　　处：《路基工程》2015年第4期15-19,共5页Subgrade Engineering

基　　金：国家自然科学基金(高铁联合基金项目):高速铁路软土地基沉降变形规律与控制方法研究(U1134207)

摘　　要：城市地铁车辆段整体道床区路基对工后沉降有严格要求。为保证路基沉降观测数据的可靠性,首先采用沉降观测异常数据判别方法,对沉降数据进行了预处理;根据路基沉降数据的特性,分别以支持向量机和神经网络法为核心技术构建了路基沉降预测模型,并通过工程实例详细介绍了预测方法与过程。对比分析表明:基于支持向量机和神经网络法构建的预测模型均有较好的预测精度;预测结果显示,依托工程路基沉降已基本趋于稳定,运营期不会发生较大的工后沉降,现有地基处理与路基填筑压实的施工方法是有效的。For the subgrade for monolithic track bed in car depot of urban MRT, the post-construction is subject to strict requirement. To ensure the reliability of data from settlement observation, they are preprocessed by abnormal data discrimination, then according to their characteristics, the models predicting subgrade settlement were established respectively by taking support vector machine and neural network as core technique, with the detailed description on prediction method and procedure depending on project case. The comparative analysis shows that the models based on SVM and neural network respectively have good accuracy; with the prediction result, the subgrade settlement is going to be steady basically and will not turn up after the construction, thus the current construction technology for foundation treatment and subgrade filling and compaction is effective.

关 键 词：路基沉降 沉降预测 支持向量机 神经网络 

分 类 号：TU433[建筑科学—岩土工程]