Abstract: In order to improve the generalization performance and prediction accuracy in the prediction of rock burst risk rating, the number of neurons in the hidden layer and the excitation function of the ELM were optimized by using grid search method with 10-fold cross-validation. Then a prediction model was build with the optimized parameters. 26 groups of 36 groups of actual measured data form Yanshitai coal mine were used to train the model and the rest 10 groups of data to test it. The result shows that the correct recognition rate by the trained model reaches 84.615%using 10-fold cross-validation, which is better than Naive Bayes's 76.92% andAdaboostM1's 61.54%, and the prediction accuracy by the trained model for the rest 10 groups of data is 90%, which is better than Naive Bayes andAdaboostM1's 80%.

Key words: rock burst, risk rating prediction, extreme learning machine(ELM), grid search method, 10-fold cross-validation