基于主成分分析-相关向量机的高速公路路基沉降量预测

Settlement Prediction of Highway Subgrade with Reterance Vector Machine Based on Principal Component Analysis

为解决高速公路路基沉降量难以获取的难题,提出一种基于主成分分析(principal compohent analysis,PCA)的相关向量机(relevance vector machine,RVM)路基沉降量预测方法。通过主成分分析法将多个易获取的土体常规物理参数降维成少数且独立的变量,借助相关向量机模型反映路基沉降量与4个主成分变量之间的非线性映射关系,建立基于PCA-RVM的高速公路路基沉降量预测模型。将该模型应用于工程实例,在同样学习样本情况下与4种神经网络预测模型对比分析,结果表明:PCA-RVM预测模型通过分析各因素的相关性与贡献率,将多个影响因素合理化为少数主成分变量,在信息筛选方面明显优于其余4种模型;各模型预测结果显示,在路基沉降量预测结果的相对误差及均方差方面,PCA-RVM预测模型均占据较大优势。PCA-RVM模型具有精度高、离散性小、可靠度高等优点,为高速公路路基沉降量预测提供了一种新方法。

To solve the difficulty in predicting the settlement of highway subgrade,a new method with the relevance vector machine(RVM)was proposed based on principal component analysis(PCA).The dimension of easily-available conventional physical parameters for soil was reduced to a few independent variables using PCA method.RVM was used to reflect the nonlinear mapping relationship between the subgrade settlement and the four principal component variables,and then the PCA-RVM model for predicting the highway subgrade settlement was established.The proposed model was applied in an engineering project and compared with four neural network prediction models using the same learning samples.Results show that the PCA-RVM model was superior to the other four models in information screening,because it reasonably reduces many influencing factors to few principal component variables by analyzing the correlation and contribution rate of each factor.The prediction results of each model show that the PCA-RVM model had more advantages for the relative error and mean square error.Therefore,the PCA-RVM model features the merits of high precision,small dispersion,and high reliability,and provides a new method for predicting the settlement of highway subgrade.