为进一步提高岩石爆破块度预测效果，利用多个矿山的岩石爆破统计数据，通过优化极端梯度提升模型（extreme gradient boosting, XGBoost）超参数，建立一种基于随机森林（random forest, RF）特征选择的麻雀搜索算法（sparrow search algorithm, SSA）优化XGBoost爆破块度预测模型。利用麻雀搜索算法对XGBoost模型决策树数量、决策树最大深度、学习率3个核心超参数进行优化以提高运行效率；利用随机森林对输入特征进行筛选，并将优化后的特征集输入预测模型。结果表明：经特征集优化的模型，爆破块度预测效果整体上更加逼近实际值，且预测结果的可决系数（R-squared, R²）、均方根误差（root mean square error, RMSE）和平均绝对误差（mean absolute error, MAE）分别为0.954、0.026和0.020，相较于BP（back propagation）神经网络、随机森林和XGBoost模型的效果更优，在实际应用中更具适用性，能为爆破参数设计和优化提供借鉴。

In order to further improve the predictive effect of rock blasting fragmentation, the blasting statistical data of several mines was used to build a prediction model, which was based on feature selection by random forest and the XGBoost regression prediction model optimized by the sparrow search algorithm. Aiming at improving the operating efficiency of the XGBoost regression prediction model, the sparrow search algorithm(SSA) was used to optimize their three core hyperparameters, including the number trees, the max depth and the learning rate. The input features selected by random forest were input into the model. The prediction effect of blasting fragmentation is closer to the actual value, and the R-squared (R²), the root mean square error (RMSE) and the mean absolute error (MAE) of the prediction results are 0.954, 0.026 and 0.020. Compared with the back propagation (BP) neural network, the random forest and the XGBoost model, the proposed model is better and more applicable. It is concluded that the proposed model is more adaptive in practical application, and can provide reference for the design and optimization of blasting parameters.