Flyrock is one of the most important environmental and hazardous issues in mine blasting, which can affect equipment and people, and may lead to fatal accidents. Therefore, prediction and minimization of this phenomenon are crucial objectives of many rock removal projects. This study is aimed to predict the flyrock distance with the use of machine learning techniques. The most effective parameters of flyrock were measured during blasting operations in six mines. In total, 262 data samples of blasting operations were accurately measured and used for approximation purposes. Then, flyrock was evaluated and estimated using three machine learning methods: principle component regression (PCR), support vector regression (SVR), and multivariate adaptive regression splines (MARS). Many models of PCR, SVR, and MARS were constructed for the flyrock distance prediction. The modeling process of each method is elaborated separately in a way to be used by other researchers. The most important parameters affecting these models were assessed to obtain the best performance for the developed models. Eventually, a preferable model of each machine learning technique was used for comparison purposes. According to the used performance indices, coefficient of determination (R²), and root mean square error, the SVR model showed a better performance capacity in predicting flyrock distance compared with the other proposed models. Thus, the SVR prediction model can be used to accurately predict flyrock distance, thereby properly determining the blast safety area. Additionally, the SVR model was optimized by new optimization algorithm namely gray wolf optimization (GWO) for minimizing the flyrock resulting from blasting operation. By developing optimization technique of GWO, the value of flyrock can be decreased 4% compared with the minimum flyrock distance.

飞石是矿山爆破中最重要的环境和危险问题之一，它可能影响设备和人员，并可能导致致命事故。因此，对该现象的预测和最小化是许多岩石清除项目的关键目标。这项研究旨在使用机器学习技术来预测飞石距离。在六个矿山的爆破作业中测量了飞石的最有效参数。总共精确测量了262个喷砂作业的数据样本，并将其用于近似目的。然后，使用三种机器学习方法对飞石进行评估和估计：主成分回归（PCR），支持向量回归（SVR）和多元自适应回归样条（MARS）。PCR，SVR，建立了MARS和MARS进行飞石距离预测。每种方法的建模过程都以其他研究者可以使用的方式分别进行了详细阐述。对影响这些模型的最重要参数进行了评估，以获得开发模型的最佳性能。最终，将每种机器学习技术的首选模型用于比较目的。根据使用的性能指标，确定系数（R ²）和均方根误差，与其他提出的模型相比，SVR模型在预测飞石距离方面表现出更好的性能。因此，SVR预测模型可用于准确预测飞石距离，从而正确确定爆炸安全区域。此外，通过新的优化算法（即灰狼优化（GWO））对SVR模型进行了优化，以使爆破操作产生的飞石最小化。通过开发GWO优化技术，与最小飞石距离相比，飞石值可减少4％。