Abstract: In order to improve the accuracy of risk prediction for rock burst, a prediction model based on preprocessing for AFOA optimizing ELM was proposed. Taking Yanshitai coal mine in Chongqing as an example, 10 factors influencing rock burst risk were selected as characteristics of the model. Taking part of the measured data as sample data sets and preprocessing, a balanced data set was constructed by using SMOTE, and the feature dimension was reduced by using GRA and FA. According to the fruit fly optimization algorithm(FOA), AFOA was constructed by introducing jump off variable and classification accuracy variance variable. AFOA was used to optimize the input layer weights and hidden layer thresholds of ELM, and a risk prediction model for rock burst was constructed. The preprocessing sample data were trained, predicted and compared with that by other models. The results show that the preprocessing of data set can significantly improve the prediction effect of AFOA-ELM model, and that the prediction accuracy of the proposed risk prediction model is 93.75%, the mean square error is 6.25%, and the prediction accuracy is significantly better than other comparison models.

Key words: rock burst, risk prediction, synthetic minority oversampling technique(SMOTE), factor analysis(FA), grey relational analysis(GRA), extreme learning machine(ELM), ameliorated fruit fly optimization algorithm(AFOA)