Abstract: In order to evaluate the risk of coal floor groundwater bursting quickly and accurately, quantities such as water pressre and excavation height were identified as factors influencing the bursting, gray relative correlations between the factors were used for covariance matrix of GSPCA, and gray principal components were obtained with nonoverlapping informations by GSPCA. A GSPCA-LSSVM evaluation model was built, which took the gray principal components as inputs and risks of coal floor groundwater bursting as outputs. The GSPCA-LSSVM model was trained through twenty groups of learning samples, and verified by the re-substitution method. The trained model was used to evaluate five groups of test samples. The results show that more than 91.97% of information of original factors has been extracted by GSPCA, which considers the incompletenes, that the redundant information and computation complexity have been reduced significantly, that the evaluation results obtained by using the model accord with the actual situation basically, and that the model could be used to evaluate the risk of coal floor groundwater bursting effectively.

Key words: risk assessment, water inrush, grey system(GS), principal component analysis(PCA), least square support vector machine(LSSVM)