Developing a hybrid model of Jaya algorithm-based extreme gradient boosting machine to estimate blast-induced ground vibrations

• Published in: International journal of rock mechanics and mining sciences (Oxford, England : 1997) Vol. 145; p. 104856
• Main Authors: Zhou, Jian, Qiu, Yingui, Khandelwal, Manoj, Zhu, Shuangli, Zhang, Xiliang
• Format: Journal Article
• Language: English
• Published: Berlin Elsevier Ltd 01.09.2021; Elsevier BV
• Subjects: Algorithms; Artificial neural networks; Blasting; Dependent variables; Empirical analysis; Ground motion; Ground vibration; Independent variables; Jaya algorithm; Learning algorithms; Machine learning; Mathematical models; Model accuracy; Neural networks; Optimization; Parameters; PPV; Root-mean-square errors; SHAP; Vibration; Vibrations; XGBoost; Jaya algorithm; SHAP; Blasting; XGBoost; Ground vibration; PPV
• ISSN: 1365-1609; 1873-4545; 1873-4545
• DOI: 10.1016/j.ijrmms.2021.104856

Blasting is still being considered to be one the most important applicable alternatives for conventional excavations. Ground vibration generated due to blasting is an undesirable phenomenon which is harmful for the nearby structures and should be prevented. In this regard, a novel intelligent approach for predicting blast-induced PPV was developed. The distinctive Jaya algorithm and high efficient extreme gradient boosting machine (XGBoost) were applied to obtain the goal, called the Jaya-XGBoost model. Accordingly, 150 sets of data composed of 13 controllable and uncontrollable parameters are chosen as input independent variables and the measured peak particle velocity (PPV) is chosen as an output dependent variable. Also, the Jaya algorithm was used for optimization of hyper-parameters of XGBoost. Additionally, six empirical models and several machine learning models such as XGBoost, random forest, AdaBoost, artificial neural network and Bagging were also considered and applied for comparison of the proposed Jaya-XGBoost model. Accuracy criteria including determination coefficient (R2), root-mean-square error (RMSE), mean absolute error (MAE), and the variance accounted for (VAF) were used for the assessment of models. For this study, 150 blasting operations were analyzed. Also, the Shapley Additive Explanations (SHAP) method is used to interpret the importance of features and their contribution to PPV prediction. Findings reveal that the proposed Jaya-XGBoost emerged as the most reliable model in contrast to other machine learning models and traditional empirical models. This study may be helpful to mining researchers and engineers who use intelligent machine learning algorithms to predict blast-induced ground vibration.