Evaluation of Various Rockburst Prediction Techniques Using Neelum Jhelum Hydropower Project’s Rockburst Data

Abstract

Rock excavation of underground spaces is extensively used in developing countries to construct hydroelectric-power plants, pumping stations, waterways, railway, and traffic tunnels, and many more of such projects. Rock excavation is mostly carried out either by tunnel boring machine (TBM) or by drill and blast method. Excavation of large volume of rock by produces disturbs the surrounding rockmass, resulting in rockbursts. In this research, Empirical correlations to predict the Rock Burst intensity were compared with the Machine Learning approaches. The database consisted of 262 records of rockburst for training and fourteen for validation. The selected parameters to develop the models were Depth, Compressive Strength, Tangential Strength, and Stress Concentration Factor (SCF). Out of 262 records, 197 (75%) records were used to train the models and 65 (25%) records were used to evaluate the models in Machine Learning models. Fourteen records from NJHPP were used for validation. The performance of the developed models was checked for accuracy of predicting the rockburst intensity. It was observed that the empirical models performed the worst with the maximum accuracy of 21.3% from Chinese Code. The results obtained from SVM outperformed all other models with the accuracy of 92.8%. The accurate classification of rockburst risks before underground excavation can provide reference for designing tunnel support systems, roadway supports, arranging destress boreholes and deploying monitoring systems. The machine learning method can be easily extended for rock burst classification problems in underground engineering. Future work should focus on enlarging the current database to improve the generalization ability of the ensemble model.