桩横向受荷承载力预测的相关向量机模型

Relevance Vector Machine Model for Predicting Lateral Load Bearing Capacity of Piles

为解决桩基建设过程中桩横向受荷承载力难以获得的难题,提出一种基于相关向量机(Relevance Vector Machine,RVM)的桩横向受荷承载力预测模型。通过对学习样本的训练建立荷载的偏心距、桩径、土的不排水抗剪强度、桩入土深度等4个主要影响因素与桩横向受荷承载力之间的非线性映射关系,能够对仅已知影响因素的新样本进行精准预测。将该模型进行实例应用并与BP神经网络模型预测的结果进行对比,在相同的影响因素数据样本条件下,RVM模型预测精度更高、离散性更小;利用置信区间对RVM模型预测的数据进行可靠性检验,预测值均分布在90%、95%及97%置信区间内,通过相关系数确定实际值和预测值的拟合度为0.979 1,进一步验证了RVM模型预测的可靠性,具有较好的推广价值。

In order to solve the problem that the pile lateral bearing capacity is difficult to obtain in the process of pile foundation construction,a prediction model of lateral load bearing capacity of piles was proposed based on the Relevance Vector Machine(RVM).Through the training of the learning samples,the nonlinear mapping relationship between the four major influencing factors(i. e.,load eccentricity,pile diameter,soil undrained shear strength,and pile depth)and the lateral load bearing capacity of the pile was established,so as to accurately predict the new samples with only known influencing factors. The model was applied to an example and compared with the results predicted by the BP neural network model. The results showed that under the same influence factor data sample condition,the RVM model had higher prediction accuracy and less discreteness. The confidence interval was used to verify the reliability of the data predicted by the RVM model,the predicted values were distributed within the 90%,95%,and 97% confidence intervals,and the correlation coefficient between the actual value and the predicted value was 0. 979 1,which further verified the reliability of the RVM model prediction and its good promotion value.