PSO(particle swarm optimization), BP (back propagation neural network), and SVM (support vector machine) were combined to improve the accuracy of prediction and early warning. According to the prediction results of PSO-BP, the SVMs optimized by particle swarm optimization algorithm were used to give early warning of gas concentration exceeding the standard, and the risk of gas exceeding the standard was predicted and classified to different degrees. The results show that the RMSE, MAE and R² indexes of the PSO-BP prediction model reflect well, the BP prediction effect is significantly improved after PSO optimization, the early warning accuracy of the SVM model is high, and the generalization ability is strong, which meets the practical application needs of gas concentration overrun control in the coal mining process.