基于相关向量机的富水砂层渣土改良试验效果预测

Prediction of Slag Improvement in Water-Rich Sand Layer Based on Relevance Vector Machine

为了精准、高效获得富水砂层渣土改良试验效果,以渗透系数、内摩擦角(改良前)、电阻率、泡沫剂浓度和膨润土浓度为输入变量,坍落度、渗透系数、内摩擦角(改良后)为输出变量,并基于24组富水砂层渣土改良数据,建立相关向量机(relevance vector machine,RVM)预测模型,将预测结果与反向传播(back propagation,BP)神经网络模型进行对比分析。结果表明:RVM模型在预测坍落度、渗透系数和内摩擦角时的平均误差分别为0.73%、0.38%和2.24%,优于BP神经网络模型的1.76%、4.53%和3.60%;通过计算皮尔逊相关系数,可知RVM预测值与实测值对应坍落度、渗透系数、内摩擦角的相关系数r分别为0.9999、0.9993和0.9878,说明拟合程度极高。由此可见,RVM模型具有预测精度高、可靠性高的优点,为富水砂层渣土改良试验效果的预测提供一种新方法。

In order to accurately and efficiently obtain the effect of the water-rich sand layer residue improvement test,permeability,internal friction angle(before modification),resistivity,foam agent concentration and bentonite concentration were used as input variables,and slump,permeability and internal friction angle(after modification)were used as output variables,a relevance vector machine(RVM)prediction model was developed based on 24 sets of water-rich sand layer residue improvement data.The prediction results and the BP neural network model were analyzed.The results show that the average errors of the RVM model in predicting slump,permeability coefficient and internal friction angle are 0.73%,0.38%and 2.24%,respectively,which are superior to 1.76%,4.53%and 3.60%of the BP neural network model.The RVM predicted and measured values corresponding to slump,permeability coefficient,and internal friction angle were calculated by Pearson correlation coefficient,and the known correlation coefficients r are 0.9999,0.9993,and 0.9878,respectively,indicating that the fitting degree is extremely high.Therefore,the RVM model features the merits of high precision and high reliability,and provides a new method for predicting the effect of water-rich sand layer residue improvement test.