Abstract: Abstract: To predict the blasting effect of rock roadway more accurately, to improve blasting efficiency and reduce production costs, the blasting data in the existing rock blasting database is used. Select 6 factors of total charge, roadway cross-sectional area, blast hole depth, cut eye charge, auxiliary eye charge, and peripheral eye charge as input, and construct a grid search Cross Validation-Support Vector Machine (Grid Search CV-SVM) regression prediction model. Use average absolute error and correlation coefficient as evaluation indicators to predict the unit consumption of explosives. SVM uses radial basis kernel function (Rbf), polynomial kernel function (Poly), and linear kernel function (Linear) to compare three kernel functions with each other, and introduces Random Forest (Random Forest) regression algorithm as the control group. The results show that the SVR-Rbf group performed the best. The correlation coefficients in the database and the actual case prediction of the Gubei coal mine were both about 0.95, and the average absolute error was at least about twice as small as the other groups. The performance of the Random Forest prediction group is not as good as the three SVM prediction groups，and the optimal model is applied to the prediction of the unit consumption of explosives for rock tunnel blasting in Gubei Mine, and the effect is good. The established Grid Search CV-SVM prediction model is an excellent method to predict the effect of rock roadway blasting.