A Statistics-Based Projection Pursuit Model for Prediction of Rock Burst Grades

Projection pursuit (PP) models have been successfully applied into prediction of rock burst grades. However, the modeling data are generated randomly from rock burst grade criterion, and consequently, these PP models are essentially the model-based implementation of rock burst grade criteria. Here we used statistical data from 108 rock burst samples to build PP models, so the predictive results were independent on rock burst grade criteria. Then Logistic models were used to express the function relationship between projected data and empirical grades. The trend relationship between indicators and rock burst grades was analyzed by odds ratio. Results show that the statistics-based PP model using the combination of sigma(theta), sigma(theta)/sigma(c), W-et is the optimal one, with highest prediction ability and can be used to predict underground engineering rock burst.