煤炭技术

传统边坡稳定性分析采用静态安全系数，不能适应露天煤矿边坡动态风险管理需求，将边坡变形实时监测数据和人工智能技术结合，进行边坡实测变形曲线趋势预测分析。选取3种典型露天煤矿边坡变形（周期型、稳定型和突变型）曲线，分别采用机器学习理论中的季节自回归积分滑动平均模型（SARIMA）、支持向量回归模型（SVR）和长短期记忆模型（LSTM），对实测边坡变形数据进行了学习训练，通过不断迭代优化模型参数，3种模型测试集和预测集的拟合精度分别达到0.87,0.92和0.97，取得了良好的训练效果。采用训练好的学习模型对未来变形曲线的趋势进行了预测，结果表明：SARIMA适合处理与季节相关，具有固定周期的边坡中长期变形曲线，SVR适合处理减速或匀速蠕变等稳定型边坡变形曲线，LSTM则适合处理加速蠕变的突变型边坡变形曲线，可用于临滑预警。机器学习方法为实现露天矿边坡“监测—预测—预警—响应”的闭环管理提供了一种有效的手段。

As the conventional static factor of safety can no longer satisfy the demands of dynamic risk management for open-pit coal-mine slopes, real-time deformation monitoring data and artificialintelligence techniques are integrated to forecast measured displacement-time curves. Three typical deformation patterns-periodic, stable and abrupt-observed in open-pit slopes are selected, seasonal autoregressive integrated moving-average(SARIMA), support-vector regression(SVR) and long shortterm memory(LSTM) algorithms are used to train the corresponding field monitoring datasets. Iterative parameter refinement yields test-to-prediction fit accuracies of 0.87, 0.95 and 0.92, respectively,indicating robust model training. The optimized models are then used to predict future deformation trends. Results show that SARIMA is best suited for medium-to long-term displacement curves that carry a fixed, season-related periodicity; SVR is well-suited for stable-type slope deformation curves that exhibit decelerating or steady-state creep; and LSTM can capture accelerating-creep signals that precede slope failure, enabling imminent-slide alarms. Machine-learning approaches thus provide an effective tool for closing the “monitor-forecast-warn-respond” loop in open-pit slope management.