Rockburst Prediction Based on Particle Swarm Optimization and Machine Learning Algorithm

Abstract

Rockburst is a complex dynamic instability failure phenomenon when excavating in high geo-stress rock mass. Timely and effective prediction of rockburst is an important guarantee for the safe and efficient construction of deep underground engineering. A total of 191 rockburst engineering cases which considering 11 factors was sorted out, and the three main influencing factors are selected as the predictor of rockburst through correlation analysis. The rockburst prediction model was established based on BP (back propagation) neural network, probabilistic neural network (PNN), and support vector machine (SVM), and particle swarm optimization (PSO) was used to optimize model parameters. The intensity classification of rockburst was predicted, and the prediction effects of the three models before and after optimization were evaluated from the three aspects of accuracy, stability and time-consuming. The results show that the established prediction models have a good effect on rockburst prediction. Among the three optimized machine learning models, the PSO-PNN model has the best prediction effect, with a prediction rate of 86.96% for rockburst intensity classification. Then a rockburst prediction system is developed based on PSO-PNN.