Productivity in opencast mining, particularly in drill-blast (DB) and surface miner (SM) operations, is crucial for optimizing efficiency and reducing costs. These operations are directly affected by fragmentation, which in turn impacts equipment utilization, loading cycle times, and downstream operations. This study analyzed field data such as rock properties, machine parameters, blast design results, and post-blast fragmentation size (0.15–0.82 m), with 0.45 m identified as the optimal fragmentation size for a 12 m³ shovel bucket. Traditional productivity assessments often use simplistic models that fail to capture the complexities of mining operations. To address this, an explainable machine learning (ML) model was developed, integrating fragmentation size, rock and machine parameters, and geometric factors to evaluate DB and SM operations in opencast coal mines. Various ML techniques, such as artificial neural network (ANN), random forest regression (RFR), gradient boosting regressor (GBT), and support vector regression (SVR), were employed to analyze these parameters. Among these, the RFR model demonstrated the highest accuracy, with a coefficients of determination (R²) of 99.5% for training and 99.2% for testing in DB datasets, and 99.9% for training and 99.5% for testing in SM datasets. Furthermore, the RFR model had the lowest root mean square error, mean absolute error, and mean absolute percentage error of 10.35, 4.788, and 2.1% for DB training datasets, and 5.53, 1.75, and 1.5% for SM training datasets, respectively, underscoring its superior performance. Using SHAP (Shapley Additive exPlanations), the study identified key productivity drivers: SM cycle time, diesel consumption, and coal face length. Fragmentation size, resulting from blasting, was also found to influence shovel efficiency and overall productivity significantly. This paper highlights the effectiveness of ensemble ML models in predicting and analyzing complex productivity dynamics in opencast mining.

露天采矿的生产率，尤其是爆破 （DB） 和露天采矿 （SM） 作业，对于优化效率和降低成本至关重要。这些操作直接受到碎片化的影响，进而影响设备利用率、装载周期时间和下游操作。本研究分析了现场数据，例如岩石特性、机器参数、爆破设计结果和爆破后碎裂尺寸 （0.15-0.82 m），其中 0.45 m 确定为 12 m³ 铲斗的最佳碎裂尺寸。传统的生产力评估通常使用简单的模型，无法捕捉采矿作业的复杂性。为了解决这个问题，开发了一个可解释的机器学习 （ML） 模型，该模型整合了碎片大小、岩石和机器参数以及几何因素，以评估露天煤矿中的 DB 和 SM 操作。采用了各种 ML 技术，例如人工神经网络 （ANN）、随机森林回归 （RFR）、梯度提升回归器 （GBT） 和支持向量回归 （SVR），来分析这些参数。其中，RFR 模型表现出最高的准确性，在 DB 数据集中训练的决定系数 （R 2） 为 99.5%，在 DB 数据集中测试的决定系数为 99.2%，在 SM 数据集中训练的决定系数 （R²） 为 99.9%，在 SM 数据集中测试的决定系数为 99.5%。此外，RFR 模型在 DB 训练数据集中具有最低的均方根误差、平均绝对误差和平均绝对百分比误差，分别为 10.35、4.788 和 2.1%，在 SM 训练数据集中具有 5.53、1.75 和 1.5%，凸显了其卓越的性能。使用 SHAP（Shapley 添加剂解释），该研究确定了关键的生产力驱动因素：SM 循环时间、柴油消耗和采煤工作面长度。 爆破产生的碎片尺寸也对铲子效率和整体生产率有显著影响。本白皮书重点介绍了集成 ML 模型在预测和分析露天采矿中复杂的生产力动态方面的有效性。