基于机器学习的边坡稳定性分析

THE SLOPE STABILITY ANALYSIS BASED ON MACHINE LEARNING

为更加快速精确的分析边坡的稳定状态,文中结合浙江北部441个边坡案例,选取了重度、黏聚力、内摩擦角、坡角、坡高、孔隙压力比6个特征参数,建立数据集。通过在预测边坡稳定状态中有良好性能表现的支持向量机、BP神经网络、随机森林、GA-BP以及PSO-BP模型进行预测精度比较,获得样本下精度最高的算法模型。研究表明基于混淆矩阵以及AUC指标进行模型精度评价,PSO-BP具有误差小以及稳定性强的优点,为后续更加快速分析边坡稳定状态提供一种新的思路。

441 slope cases in northern Zhejiang province are taken to analyze the stable state of the slope more quickly and accurately by establishing the data set in terms of the relative density,cohesion,internal friction angle,slope angle,slope height,and pore pressure ratio.By comparing the prediction accuracy of support vector machines,BP neural networks,random forests,GA-BP,and PSO-BP models,which have shown good performance in predict-ing the stability of slopes,the algorithm model with the highest accuracy under this sample is obtained.Research shows that,based on the confusion matrix and AUC metrics for model accuracy evaluation,PSO-BP has the advan-tages of small error and strong stability,providing a new approach for faster analysis of slope stability in the future.