露天煤矿抛掷爆破爆堆形态是影响抛掷爆破–拉斗铲倒堆工艺系统作业效率和生产成本的重要因素。为了提高露天煤矿抛掷爆破爆堆形态预测的精确度，进一步优化抛掷爆破设计并降低矿山抛掷爆破剥离成本，依据黑岱沟露天煤矿抛掷爆破实测数据进行实例分析，采用熵值法–灰色关联度方法研究抛掷爆破效果影响指标与最远抛掷距离、松散系数和有效抛掷率的权重及关联度，孔距和剖面宽与抛掷爆破效果评价指标的关联程度较低，选取最小抵抗线、排距、炸药单耗、台阶高度、采空区下口宽、坡面角、采空区上口宽作为预测模型输入参数。引入高斯(Guass)分布模型模拟爆堆剖面曲线，通过1～8阶高斯分布模型对爆堆剖面曲线模拟分析，确定阶数为5时模拟精度与效率达到最优，利用训练完成的HHO–LSSVM预测5阶Guass分布模型15个参数作为预测输出的爆堆形态的控制参数，并采用Gauss分布模型结合HHO–LSSVM算法预测爆堆形态，与LSSVM、粒子群算法(PSO)优化最小二乘支持向量机和遗传算法(GA)优化BP神经网络模型进行精度对比，同时将本文所建立模型的预测爆堆形态与真实爆堆形态进行对比。结果表明：利用5阶高斯分布模拟的爆堆曲线误差平方和(S)参数趋于稳定为25.69，决定系数(R²)与调整后的决定系数($R_{mathrm{A}}^2 $)分别为0.999 2和0.999 0，均方根误差(R)为0.514 6；HHO–LSSVM对于5阶Gauss分布控制参数的预测误差大部分在1%左右，且误差没有超过5%，与LSSVM、PSO、GA–BP算法模型相比精度较高；以E5–8、E5–9及E5–10剖面为例，预测爆堆形态与真实爆堆形态误差R²和均方根误差R分别为0.998 7和0.614 2、0.999 2和0.493 1、0.999 2和0.505 2，预测爆堆形态接近真实爆堆形态。

The shape of the blast casting muck pile in surface coal mines is an important factor that affects the operational efficiency and cost of the blast casting pulling shovel stacking process system. In order to improve the accuracy of predicting the shape of blast casting muck piles in surface coal mines, further optimize the design of blast casting, and reduce the cost of blast casting stripping, based on the measured data of blast casting in the Heidaigou surface coal mine, an example analysis was conducted. The entropy method grey correlation method was used to study the weight and correlation between the impact indicators of blast casting effect and the farthest throwing distance, looseness coefficient, and effective throwing rate. The correlation between hole spacing, section width and the evaluation indicators of blast casting effect is relatively low. The minimum resistance line, row spacing, explosive consumption, step height, lower mouth width of goaf, slope angle, and upper mouth width of goaf were selected as input parameters for the prediction model. The Gaussian distribution model was introduced to simulate the profile curve of the blast casting muck pile. By using a 1–8th order Gaussian distribution model to simulate and analyze the detonation profile curve, it was determined that the simulation accuracy and efficiency reach an optimal level when the order is 5. The trained HHO–LSSVM was used to predict 15 parameters of the 5th order Gaussian distribution model, which were used as the control parameters for predicting the output detonation shape. And the Gaussian distribution model combined with the HHO–LSSVM algorithm was used to predict the shape of the blast casting muck pile, compare the accuracy of the LSSVM, Particle Swarm Optimization (PSO) optimized Least Squares Support Vector Machine, and Genetic Algorithm (GA) optimized BP neural network models, at the same time, compare the predicted blast casting muck pile morphology with the actual blasting muck pile morphology. The results show that the sum of squared errors (S) parameter of the blasting muck pile morphology curve simulated using a 5th order Gaussian distribution tends to stabilize at 25.69, and the coefficient of determination (R²) and adjusted coefficient of determination ($R_{mathrm{A}}^2 $) are 0.999 2 and 0.999 0, respectively. The root mean square error (R) is 0.514 6. The prediction error of the HHO–LSSVM for the 5th order Gaussian distribution control parameters is mostly around 1%, and the error does not exceed 5%, compared with the LSSVM, PSO, and GA–BP algorithm models, the accuracy is higher. Taking the profiles E5–8, E5–9, and E5–10 as examples, the errors R² and R between the predicted and actual detonation morphology are 0.998 7 and 0.614 2, 0.999 2 and 0.493 1, 0.999 2 and 0.505 2, respectively, the predicted shape of the blasting muck pile is close to the actual shape of the blasting muck pile morphology.