Predicting Blast-Induced Ground Vibration in Open-Pit Mines Using Different Nature-Inspired Optimization Algorithms and Deep Neural Network

• Published in: Natural resources research (New York, N.Y.) N.Y.), 2021-12, Vol.30 (6), p.4695-4717
• Main Authors: Nguyen, Hoang, Bui, Xuan-Nam, Tran, Quang-Hieu, Nguyen, Dinh-An, Hoa, Le Thi Thu, Le, Qui-Thao, Giang, Le Thi Huong
• Format: Article
• Language: English
• Subjects: Accuracy; Algorithms; Artificial neural networks; Blasting (explosive); Chemistry and Earth Sciences; Coal mines; Computer Science; Deep learning; Deformation effects; Earth and Environmental Science; Earth Sciences; Error reduction; Explosions; Flow stability; Fossil Fuels (incl. Carbon Capture); Geography; Global optimization; Groundwater; Machine learning; Mathematical Modeling and Industrial Mathematics; Mineral Resources; Mines; Neural networks; Optimization algorithms; Original Paper; Performance indices; Physics; Slope stability; Statistics for Engineering; Structural stability; Sustainable Development; Vibration
• ISSN: 1520-7439; 1573-8981
• DOI: 10.1007/s11053-021-09896-4

Blast-induced ground vibration (GV) is a hazardous phenomenon in open-pit mines, and it has unquestionable effects, such as slope instability, deformation of structures, and changing the flow direction of groundwater. Therefore, many studies in recent years have focused on the accurate prediction and control of GV in open-pit mines. In this study, three intelligent hybrid models were examined for predicting GV based on different nature-inspired optimization algorithms and deep neural networks. Accordingly, a deep neural network (DNN) was developed for predicting GV under the enhancement of deep learning techniques. Subsequently, aiming at improving the accuracy and reducing the error of the DNN model in terms of the prediction of blast-induced GVs, three optimization algorithms based on the behaviors of whale, Harris hawks, and particle swarm in nature (abbreviated as WOA, HHOA, and PSOA, respectively) were considered and applied, namely HHOA–DNN, WOA–DNN, and PSOA–DNN, respectively. The results were then compared with those of the conventional DNN model through various performance indices; 229 blasting events in an open-pit coal mine in Vietnam were processed for this aim. Finally, it was found that the proposed intelligent hybrid models outperform the DNN model with deep learning techniques, although it is a state-of-the-art model that has been recommended and claimed by previous researchers. In particular, HHOA, WOA, and PSOA (with global optimization) further improved the accuracy of the DNN model by 1–2%. Of those, the HHOA–DNN model provided the highest performance with a mean-squared-error of 2.361, root mean squared error of 1.537, mean absolute percentage error of 0.123, variance accounted for of 93.015, and coefficient determination of 0.930 on the testing dataset. The findings also revealed that the explosive charge per blast, monitoring distance, and time delay per each blasting group are necessary parameters for predicting GV.