Aiming at the problems of long time-consuming, high cost and difficulty in obtaining effective experimental data in traditional research methods of ignition and explosion mechanism, the experimental results of ignition and explosion in semi-enclosed space with barrier conditions were predicted by using a method combining data-driven with physical experiment. Based on the experimental data of combustion and explosion, a BP neural network prediction model in basis of Adam optimization algorithm was developed, and the optimal configuration of the number of neurons in the hidden layer was realized by sensitivity analysis. Taking the flame speed and maximum explosion pressure from the experiment as the characteristic sample data for training and testing, and the performance of the prediction model was evaluated by R² (determination coefficient) evaluation index, and compared with RSM model and ridge regression model. The results show that the flame velocity and the maximum detonation pressure predicted by Adam-BP model are 30% and 16% higher respectively than those predicted by RSM model, and is 10% and 8% higher than those of the ridge regression model with better robustness. The test results show that the accuracy of Adam-BP model is more than 95% in the prediction of the experimental results of pre-mixed combustible gases.Therefore, Adam-BP model is suitable for predicting the consequences of premixed combustible gas combustion and explosion, which can provide a fast prediction method for studying the combustion and explosion consequences of premixed combustible gas.