Rockburst is a relatively dangerous engineering geological disaster in underground hard rock engineering constructed in high geostress area.Due to the re-distribution of the stress in surrounding rocks during the excavation of underground engineering，the elastic strain energy is released suddenly and abruptly，causing rock fragments to eject from the rock.And then，the casualties and equipment damage are often happened，which make the rockburst become one of the worldwide difficulties in underground engineering.Therefore，the prediction of possibility of rockburst and its intensity is a problem that must be solved in underground engineering construction.For predicting rock-burst intensity effectively，a combined prediction model based on kernel principal component analysis （KPCA） of multiple types and the support vector machine （SVM） optimized by genetic algorithm or particle swarm optimization algorithm （GA/PSO） was established.According to the characteristics and causes of rockburst，rocks’ maximum tangential stress {\sigma }_{\mathrm{\theta }}，rocks’ uniaxial compressive strength {\sigma }_{\mathrm{t}}，rocks’ uniaxial tensile strength {\sigma }_{\mathrm{c}}，stress concentration coefficient S C F，rock brittleness coefficient B_{\mathrm{1}} and B_{\mathrm{2}}，and elastic energy index W_{\mathrm{e}\mathrm{t}} were chosen to form the rockburst prediction indexes system.Based on 246 groups of typical rockburst cases at home and abroad，the data were preprocessed through the principal component analysis and the principal component analysis based on linear kernel function，radial basis function （RBF） kernel function and multi-layer perceptron （MLP） kernel function.On the basis of ensuring the amount of information in the original data，2 to 4 linearly independent principal components are obtained，which reduces the correlation between the indicators and the input parameters of the SVM model，and simplifies the training process.Then input the dimensionality-reduced data into GA/PSO optimized SVM model for training and prediction.To improve classification accuracy and generalization ability of the SVM，GA/PSO were adopted to automatically determine the parameters for support vector machine，and the optimal values of parameters C and g were determined by the method of 10 fold cross validation，which avoided the blindness of manually providing parameters.In this study，220 rockburst samples were randomly selected as the training set，and the remaining 26 samples were selected as the test set.After testing，the optimal parameters，the training set and test set accuracy of the 8 combined models were obtained.The prediction accuracy of the model based on the combination of the principal component analysis method of RBF kernel function and PSO-SVM reached 92.3%，which was the optimal combination model.It demonstrated that the combined prediction model can accurately deal with the complex non-linear relationship between various factors affecting the rockburst intensity，and the model has strong engineering practicability in the prediction of rockburst intensity.

http://www.goldsci.ac.cn/article/2020/1005-2518/1005-2518-2020-28-4-575.shtml