针对传统方法预测边坡稳定性的局限性、单一模型预测准确率和性能低以及传统超参数优化方法存在的缺点等问题，在搜集统计中外276个边坡稳定性样本的坡高、坡角、容重、黏聚力、内摩擦角和孔隙压力比数据的基础上，建立了边坡稳定性预测数据库，构建一种基于贝叶斯优化（Bayesian optimization, BO）的卷积神经网络（convolutional neural networks, CNN）和长短期记忆人工神经网络（long short-term memory, LSTM）的边坡稳定性预测模型。通过贝叶斯优化调节模型中的超参数，避免过拟合，提高模型性能，并结合CNN和LSTM的优点，提高了组合模型的训练效率和泛化能力，减少了过拟合的风险。通过5项评价指标对所提模型进行性能评价并与其余5种不同的预测模型进行对比分析，结果表明：所提模型准确率、精确率、召回率、F₁分数和曲线下的面积（area under curve, AUC）分别为92.77%、95.74%、91.84%、0.937 5和0.930，相较于其他对比模型展现出更为优越的性能。将BO-CNN-LSTM模型应用到拉拉铜矿和东苗家滑坡体共9个边坡工程案例进行验证，预测结果符合工程实际，证明该模型科学可靠。

Addressing the limitations of traditional methods for predicting slope stability, the low accuracy and performance of single models, and the drawbacks of conventional hyperparameter optimization methods, a slope stability prediction database was established based on the collected statistical data of slope height, slope angle, unit weight, cohesion, internal friction angle, and pore pressure ratio from 276 slope stability samples at home and abroad. A slope stability prediction model based on Bayesian optimization (BO), convolutional neural networks (CNN), and long short-term memory artificial neural networks (LSTM) was constructed. The model hyperparameters of this composite model were optimized through BO to prevent overfitting and enhance model performance. By combining the strengths of both CNN and LSTM, the training efficiency and generalization capability of the combined model was improved and the risk of overfitting was reduced. The performance of the model presented was evaluated using five evaluation indicators and conducting a comparative analysis with five different prediction models. The results show that the accuracy, precision, recall, F₁ score, and area under curve(AUC) of the model are 92.77%, 95.74%, 91.84%, 0.937 5, and 0.930, the model presented exhibits superior performance compared to the other comparison models. The BO-CNN-LSTM model was applied to a total of nine slope engineering cases in Lala copper mine and Dongmiaojia landslide for validation, and the prediction results were consistent with the actual engineering, proving the model to be scientifically reliable.