基于PSO‑LightGBM模型的边坡稳定性预测研究

Research on Slope Stability Prediction Based on PSO‑LightGBM Model

摘要:

边坡稳定性的准确预测对降低边坡失稳风险具有重要意义。为高效且准确地预测边坡稳定性，提出了一种基于粒子群（PSO）优化轻量级梯度提升机（LightGBM）的边坡稳定性预测模型，即 PSO?LightGBM 模型。该模型首先采用粒子群算法优化 LightGBM 模型中的重要参数，在实际工程应用中，降低了 LightGBM 模型参数所产生的影响。然后采用优化后的 LightGBM 模型对边坡稳定性进行分类预测。选取 K 近邻（KNN）、支持向量机（SVM）、 LightGBM、网格搜索优化 LightGBM（GS?LightGBM）以及遗传优化 LightGBM（GA?LightGBM）作为对比模型，并采用准确率、精确率、召回率与 F1 分数作为各模型预测性能的评价指标，并通过混淆矩阵可视化各模型的分类结果。基于 PSO?LightGBM 模型的特征重要性分析，量化了各因素在边坡稳定性预测中的相对重要性。研究结果表明，在测试集上 PSO?LightGBM 模型的各项评价指标上均显著优于其他对比模型，表现出较强的分类预测性能与泛化能力。通过特征重要性分析，影响边坡稳定性的因素从大到小依次为：坡角、坡高、内聚力、内摩擦角、土体重度与孔隙水压力。本研究为边坡稳定性的准确预测提供了一种新方法，对边坡工程安全设计与风险评估具有重要参考意义。

Abstract:

The accurate prediction of slope stability is of great significance for reducing the risk of slope instability. To achieve efficient and accurate slope stability prediction, this study proposed a slope stability prediction model based on particle swarm optimization (PSO) for light gradient boosting machine (LightGBM), namely the PSO-LightGBM model. This model first used the PSO algorithm to optimize key parameters of the LightGBM model, reducing the impact of parameters of the LightGBM model in practical engineering applications. Then, the optimized LightGBM model was adopted to classify and predict the slope stability. K-nearest neighbor (KNN), support vector machine (SVM), LightGBM, grid search-optimized LightGBM (GS-LightGBM), and genetic algorithm-optimized LightGBM (GA-LightGBM) were selected as comparison models. Accuracy, precision, recall, and F1 score were used as evaluation indicators for the predictive performance of each model, and the classification results of each model were visualized through the confusion matrices. Based on the feature importance analysis of the PSO-LightGBM model, the relative importance of each factor in slope stability prediction was quantified. The results showed that the PSO-LightGBM model significantly outperformed all other comparison models across all evaluation indicators on the test set, demonstrating strong classification prediction performance and generalization ability. Through feature importance analysis, the factors affecting slope stability were ranked in descending order as follows: slope angle, slope height, cohesion, internal friction angle, soil unit weight, and pore water pressure. This study provides a new method for the accurate prediction of slope stability and offers an important reference for the safety design and risk assessment of slope engineering.