基于机器学习的边坡稳定性预测模型优化研究

【作者】 熊振涛；

【导师】 王超； 郑万成；

【作者基本信息】 昆明理工大学 ， 安全工程（专业学位）， 2023， 硕士

【摘要】 边坡工程作为岩土工程的重要课题之一,广泛应用于采矿、土木、铁路公路、水利水电等工程项目。准确高效地开展边坡稳定性预测,对边坡失稳防治具有至关重要的作用。针对目前机器学习模型在边坡稳定性预测中存在的收敛速度慢、预测精度低等问题,本文建立边坡稳定性样本数据库,采用智能优化算法对模型进行优化,分别建立三种边坡稳定性预测模型,并对预测结果进行综合分析,最后将初步优选出的模型应用于大理某建设工程项目西北侧边坡进行验证。研究成果对边坡稳定性预测及失稳防治具有一定的指导意义。(1)建立了边坡稳定性预测样本数据库。选取边坡高度H、边坡角β、重度γ、粘聚力C、内摩擦角Ф和孔隙压力比ru等6个参数作为边坡稳定性的预测指标。通过国内外文献收集边坡稳定性样本数据,去除重复值和异常值,建立了包含167组边坡工程案例的样本数据库。(2)采用三种优化算法对梯度提升树(GBDT)模型的超参数进行优化,建立三个边坡稳定性预测模型并进行性能评估,优选出最佳GBDT模型,即PSO-GBDT边坡稳定性预测模型,其预测准确率为92.33%。(3)引入三种优化算法对支持向量机(SVM)的两个超参数c和g进行优化,分别建立三种SVM预测模型并进行性能评估,优选出最佳SVM模型,即基于麻雀搜索算法优化的支持向量机(SSA-SVM)模型,其预测准确率为93.41%。(4)通过核函数将极限学习机(ELM)进行改进得到核极限学习机(KELM),基于三种策略改进鲸鱼优化算法得到全局搜寻策略的鲸鱼优化算法(GSWOA),采用GSWOA对KELM的超参数正则化系数C和核函数参数S进行优化并建立GSWOA-KELM预测模型,将预测结果与未改进的WOA-KELM模型进行对比,优选出最佳KELM预测模型即GSWOA-KELM模型,其预测准确率为95.2%。(5)将3种边坡稳定性预测的初步优选模型应用于大理某建设工程项目西北侧边坡进行验证,综合分析各模型的预测结果,最终优选出本文最佳模型为基于GSWOA-KELM的预测模型。该模型在大规模样本数据库中的预测准确率为97.78%,在小规模样本中的预测结果与实际工况完全相符。更多还原

【Abstract】 Accurate and efficient prediction of slope stability is of paramount importance in the field of slope engineering,which is a significant discipline within geotechnical engineering.Its applications span across various engineering projects such as mining,civil infrastructure,railways,highways,water management and hydropower.To address the challenges encountered with slow convergence speed and low prediction accuracy in machine learning models for slope stability prediction,this study endeavors to establish a slope stability sample database and optimize the models using intelligent optimization algorithms.Three slope stability prediction models are developed and their outcomes are subjected to comprehensive analysis.Finally,the preliminarily selected model was applied to the northwest slope of a construction project in Dali for verification.The research findings hold substantial significance for enhancing slope stability prediction accuracy and aiding in the prevention and control of slope instability.(1)A slope stability prediction sample database is established.Six parameters,including slope height(H),slope angle(β),unit weight(γ),cohesion(C),angle of internal friction(Φ),and pore water pressure ratio(ru),are selected as the prediction indicators for slope stability.By collecting slope stability sample data from domestic and international literature,removing duplicate and abnormal values,a sample database containing 167 sets of slope engineering cases is established.(2)Three optimization algorithms are used to optimize the hyperparameters of the Gradient Boosting Decision Tree(GBDT)model.Three slope stability prediction models are established and evaluated for performance.The optimal GBDT model,namely the PSO-GBDT slope stability prediction model,is selected with a prediction accuracy of92.33%.(3)Three optimization algorithms are introduced to optimize the two hyperparameters,c and g,of the Support Vector Machine(SVM).Three SVM prediction models are established and evaluated for performance.The optimal SVM model,namely the SSA-SVM model optimized by the Sparrow Search Algorithm,is selected with a prediction accuracy of 93.41%.(4)The Extreme Learning Machine(ELM)is improved by introducing kernel functions,resulting in the Kernel Extreme Learning Machine(KELM).Based on three strategies for improving the Whale Optimization Algorithm(WOA),the Global Search Strategy Whale Optimization Algorithm(GSWOA)is obtained.GSWOA is used to optimize the regularized coefficient(C)and kernel function parameter(S)of KELM,and the GSWOA-KELM prediction model is established.The prediction results are compared with the unimproved WOA-KELM model,and the optimal KELM prediction model,namely the GSWOA-KELM model,is selected with a prediction accuracy of95.2%.(5)The preliminary optimization models of three kinds of slope stability prediction were applied to the northwest slope of a construction project in Dali for verification.The prediction results of each model are comprehensively analyzed.Finally,the GSWOAKELM prediction model,based on the GSWOA-KELM model,is selected as the best model in this study.The prediction accuracy of the model in the large-scale sample database is 97.78%,and the prediction results in the small-scale samples are in complete agreement with the actual working conditions.更多还原