Indirect hazard evaluation by the prediction of backbreak distance in the open pit mine using support vector regression and chicken swarm optimization

Abstract

Listen

Translate article

Backbreak is one of the undesirable phenomena in open-pit mines and causes several adverse hazards, such as lanslide, rock falling off and bench instability. Backbreak is influenced by many factors, such as rock properties, blasting design and local geology, so it is very difficult to assess and evaluate backbreak accurately. Therefore, controlling and accurate prediction of backbreak distance are crucial tasks to reduce hazards in open-pit mines. For this, soft computing-based techniques are considered to be an effective means, as they can integrate various sophisticated factors into a function to predict and evaluate backbreak distance. So, in this study, support vector regression (SVR) based techniques and three different types of bio-inspired meta-heuristic (BIMH) algorithms, such as chicken swarm optimization (CSO), whale optimization algorithm (WOA) and seagull optimization algorithm (SOA) are used to develop backbreak distance prediction models. The support vector regression is used as a regression tool and BIMH algorithms are used to optimize the hyper-parameters in the support vector regression. Four different types of evaluation metrics are utilized to assess the model performance, namely coefficient of determination (R2), mean square error (MSE), mean absolute error (MAE) and variance account for (VAF). An integrated evaluation system is adopted to provide overall performance for each backbreak prediction scenario. It can be indicated that CSO-SVR based backbreak prediction models can procure the best comprehensive performance and also show the best calculation efficiency. Detailed results include R2, VAF, MSE and MAE equal to 0.99475, 0.034, 99.477 and 0.1553 for a testing set and 0.97450, 0.1633, 97.466, and 0.1914 for a training set which can be said to be an excellent prediction result. By doing this, the hazard risk induced by backbreak can be indirectly assessed. In addition, it is also found that some superior performance can be obtained in some evaluation metrics compared with previous studies which utilized the same backbreak dataset for prediction.