基于智能优化算法的边坡稳定性预测方法研究
Research on Slope Stability Prediction Method Based on Intelligent Optimization Algorithm
杨小平 1段生锐 1蒋力 2刘光辉3
作者信息
- 1. 桂林理工大学 信息科学与工程学院,广西 桂林 541004;桂林理工大学 广西嵌入式技术与智能系统重点实验室,广西 桂林 541004
- 2. 广西壮族自治区地质环境监测站,广西 南宁 530029
- 3. 桂林赛普电子科技有限公司,广西 桂林 541004
- 折叠
摘要
针对边坡稳定性预测中数据分析片面、模型预测精度低的问题,基于 302 个边坡案例,选取 6 个变量特征,利用麻雀搜索算法(SSA)更新BP神经网络的敏感因子,建立 SSA-BP边坡稳定性预测模型.采用混淆矩阵、受试者工作特征(ROC)曲线及曲线下面积AUC 值作为衡量指标,通过五折交叉验证法提高模型的泛化能力并与 RF、BP、SVM、PSO-BP、GA-BP和 LSTM 6 种机器学习算法进行预测效果对比.结果表明,SSA-BP模型的AUC 值、准确率和F1 分数均最高,分别为 91.90%、85.81%和 85.87%,相较于优化前的BP网络AUC值提高了 23%.经典算例证明 SSA-BP预测模型与 ABAQUS计算的安全系数相近,并可给出可靠的预测结果,为岩土工程中边坡稳定性预测提供了一种新方法.
Abstract
Aiming at the issues of biased data analysis and poor model accuracy in prediction of slope stability,six variable features are chosen based on 302 slope examples,and the BP neural network's sensitive parameters are updated by the sparrow search algorithm(SSA).The SSA-BP slope stability prediction model is established.As measurement indi-cators,the confusion matrix,the ROC curve,and the area under the curve(AUC)value were employed.The five-fold cross-validation procedure increased the model's capacity for generalization.The five machine learning algorithms of RF,BP,SVM,PSO-BP,GA-BP and LSTM were compared.The results show that the AUC value,accuracy and F1 score of the SSA-BP model were the highest,and the responding values reached 91.90%,85.81%and 85.87%,respectively,which was 23%higher than that of the BP network before optimization.The classical example proved that the SSA-BP prediction model is similar to the safety factor calculated by ABAQUS,which can give reliable prediction results.Thus,it provides a new way for slope stability prediction in geotechnical engineering.
关键词
边坡/稳定性预测/机器学习/麻雀搜索算法(SSA)/BP网络/混淆矩阵Key words
slope/stability prediction/machine learning/sparrow search algorithm/BP network/confusion matrix引用本文复制引用
基金项目
广西桂林市科技局科技项目(20220107-1)
国家高技术研究发展计划(863计划)(2013AA12210504)
广西壮族自治区科技攻关计划(AC1638012)
出版年
2024
NETL
NSTL
万方数据