Simulation of deflagration test of premixed combustible gas with WOA-BRANN

The deflagration test of premixed combustible gas under conditions of obstacles is difficult and costly. In view of the issue, a test result prediction algorithm based on whale optimization algorithm and Bayesian regularization artificial neural network is introduced. The model is built jointly by mechanism analysis and real data training. Since the volume of available data is limited, the data for training is screened first, only the data significantly deviating from calculation results（about 80% of the total data set） is selected. The configuration of hidden layers of neuron elements are optimized based on sensitivity analysis. The coefficient of determination（R²） is used to estimate the performance of the prediction. A response surface model and a BP model are constructed for comparison. Experiments show that R² of the flame velocity and pressure values from the optimized Bayesian regularization neural networks both improved 15% compared to those from BP network. They are 34% and 29% better compared to those from RSM model respectively. The fluctuation of R² decreased more than 50%. The introduction of the whale optimization algorithm further improves the robustness of the model.