Abstract: In order to evaluate the intensity level of the coal's bursting liability and to solve the difficult problem that fuzzy comprehensive evaluation method cannot distinguish the bursting liability of 8 kinds of coal samples, the Bayes discriminant analysis method was introduced for classification of coal's bursting liability. Firstly, the duration of dynamic fracture, elastic energy index, bursting energy index and uniaxial compressive strength were selected as classification indicators. Then the Bayes discriminant model was established by taking 110 groups of data of bursting liability as training samples. Finally, four dimensionless methods were used to process the original data of the evaluation index, and the corresponding discriminant model was established. The influence of dimensionless methods on the discriminant accuracy of Bayes model was also studied. The results show that the Bayes model based on normalized method has the highest accuracy, reaching 98.2%, that the classification results of 10 different engineering samples by Bayes model are in good agreement with actual situation, and that the proposed model can avoid the influence of indicator correlation on classification results of coal's bursting liability.

Key words: rockburst, bursting liability, indicator correlation, dimensionless, Bayes discriminant model, influence analysis