GA-BP Neural Network Based Meta-Model Method for Computational Fluid Dynamic Approximation

The prediction of gas diffusion concentration has practical significance. Because the process simulation is based on complex mechanism, it will not be able to be calculated in real time. Moreover, the requirements of computational fluid dynamics on computers limit its application. This paper proposes the computational fluid dynamics simulation surrogate models based on the GA-BP neural network to predict the concentration after aerosol dispersion. Considering the relevant influence parameters of time, space coordinates and concentration, two different models of input and output variables are constructed. The results reveal that when the prediction object is affected by high-dimensional complex factors, the GA-BP neural network can generate accurate prediction results. Compared with the traditional BP neural network, the prediction accuracies can be improved by 40.65% and 77.61%, respectively, which exhibits excellent performance for data prediction. The method proposed in this paper successfully verifies the computational fluid dynamics simulation of the aerosol dispersion processes, and the research has potential application value for environmental safety assessment.