Abstract:

In order to improve the prediction accuracy of rockburst intensity grade, a prediction model based on HDO and AHDPSO-ELM was proposed. Firstly, the main evaluation indexes were selected based on the analysis of the influencing factors of rockburst intensity. The HDO algorithm was used to increase the number of minority samples and balance the rockburst samples of each intensity grade. Then, based on particle swarm optimization (PSO), the adaptive population spacing and mutation operator in differential evolution algorithm (DE) were introduced to design AHDPSO. AHDPSO optimized the input layer weight and hidden layer threshold of ELM, and the rockburst grade prediction model was constructed. Finally, 301 sets of rockburst samples at home and abroad were used to train, test, and compare with other models. The results show that after improving the structure of datasets by the HDO algorithm, the overall average accuracy of rockburst prediction is increased by 11.91%, and the average prediction accuracy of each grade has been improved. The average prediction accuracy of the AHDPSO-ELM rockburst intensity prediction model based on HDO is 98.92%, and the mean square error is 0.010 8. The prediction accuracy is better than other comparison models.

Key words: Hellinger distance oversampling (HDO), adaptive hybrid differential particle swarm optimization (AHDPSO), prediction of rockburst intensity grade, extreme learning machine (ELM), rockburst samples, mutation operator, adaptive population spacing