Rockburst grade probability prediction models based on PSO parameter optimization

Rockburst is a complex dynamic hazard in underground engineering, with the characteristics of sudden, random and destructive, seriously threatening the safety of construction personnel and mechanical equipment, limiting the project schedule. This paper collects 403 groups of rockburst cases, extracts four representative rockburst prediction indicator data, and uses LOF algorithm to process the outliers in the data sets. The processed data are used to test the prediction performance of ensemble models and decision tree models. Meanwhile, Particle swarm optimization (PSO) algorithm is used to optimize the parameters of the prediction models. The results show that the prediction performance of the ERT models is better than that of the RF models and the decision tree models; the CART-ERT model has the highest prediction accuracy of 0.9375, which is better than the other eight prediction models. Finally, ADASYN algorithm is used to synthesize minority classes of samples to reduce the impact of class imbalance of rockburst samples. It can be found that after using the ADASYN algorithm to synthesize samples, the prediction performance of the models is improved. The prediction models adopted in this paper calculate the occurrence probability of rockburst of different grades, which has important guiding significance for rockburst prevention and control.