Abstract: In order to predict rockburst disaster accurately and reliably, RBFNN, PNN and GRNN prediction models based on PCA were established. Six frequently-used parameters were chosen to constitute prediction indicator system, PCA was used to eliminate correlation of indicators and reduce their dimensionality. Then, three linearly independent pivot elements were obtained, namely three comprehensive indicators Y₁, Y₂ and Y₃, which constituted input vectors of RBFNN, PNN and GRNN neural networks. The results show that predictions of three PCA neural network models are better than original RBFNN, PNN and GRNN models as they not only improve accuracy, but also shorten operation time. Moreover, according to comparison from three aspects of local accuracy, overall accuracy and operation time, these three models ranks as PCA-GRNN > PCA-PNN > PCA-RBFNN > PNN > GRNN > RBFNN from strong to weak based on their accuracy ability.

Key words: principal component analysis(PCA), radial basis function neural network(RBFNN), probabilistic neural network(PNN), generalized regression neural network(GRNN), rockburst prediction