Qi Yun, Bai Chenhao, Li Xuping, Duan Hongfei, Wang Wei, Qi Qingjie
School of Mining and Coal, Inner Mongolia University of Science and Technology, Baotou, 014010, China.
College of Coal Engineering, Shanxi Datong University, Datong, 037000, China.
Sci Rep. 2025 May 31;15(1):19133. doi: 10.1038/s41598-025-01988-y.
Slope instability is a prevalent dynamic disaster encountered in the construction of geotechnical engineering projects. Intelligent detection and early warning systems serve as crucial measures for preventing and controlling slope instability. To accurately and efficiently predict the stability state of slopes, we propose a combined model that integrates the Newton-Raphson optimization algorithm (NRBO) with an optimized extreme gradient boosting tree (XGBoost). Firstly, the primary factors influencing slope instability are thoroughly analyzed. The sample outline is standardized utilizing polar deviation, and the distribution of sample classes is balanced through the application of the Synthetic Minority Over-sampling Technique (SMOTE). Secondly, the XGBoost model is optimized by fine-tuning parameters such as maximum depth (max_depth), learning rate (learning_rate), subsample rate, column sampling rate (colsample-bytree), and minimum loss (gamma) through NRBO. The stability of the model was assessed using a ten fold cross validation method, while the prediction results were comprehensively evaluated utilizing metrics including accuracy (Acc), precision (Pre), recall (Rec), F1 score (Fs), and Cohen's Kappa coefficient (Ka). Finally, the SHAP additive interpretation method is employed to elucidate the significance and contributions of features influencing the XGBoost model. This model is subsequently applied to ten specific engineering case studies. The results show that after NRBO optimization, the optimal values for the maximum depth, learning rate, subsample proportion, column sample proportion, and minimum loss of the XGBoost model are 7, 0.8247, 0.6326, 0.6263, and 0.0758, respectively. Based on the SHAP model analysis, the main factors influencing the stability of the slopes are g, c, φ, H, and j.
边坡失稳是岩土工程项目建设中常见的动力灾害。智能检测与预警系统是防治边坡失稳的关键措施。为准确、高效地预测边坡的稳定状态，我们提出了一种将牛顿 - 拉夫逊优化算法（NRBO）与优化后的极端梯度提升树（XGBoost）相结合的模型。首先，深入分析影响边坡失稳的主要因素。利用极坐标偏差对样本轮廓进行标准化，并通过应用合成少数过采样技术（SMOTE）平衡样本类别的分布。其次，通过NRBO对最大深度（max_depth）、学习率（learning_rate）、子采样率、列采样率（colsample-bytree）和最小损失（gamma）等参数进行微调，对XGBoost模型进行优化。使用十折交叉验证方法评估模型的稳定性，同时利用准确率（Acc）、精确率（Pre）、召回率（Rec）、F1分数（Fs）和科恩卡帕系数（Ka）等指标对预测结果进行综合评估。最后，采用SHAP加性解释方法阐明影响XGBoost模型的特征的重要性和贡献。该模型随后应用于十个具体的工程案例研究。结果表明，经过NRBO优化后，XGBoost模型的最大深度、学习率、子采样比例、列采样比例和最小损失的最优值分别为7、0.8247、0.6326、0.6263和0.0758。基于SHAP模型分析，影响边坡稳定性的主要因素是g、c、φ、H和j。
