Abstract
When the indoor environment is designed by genetic algorithm (GA) and computational fluid dynamics (CFD), the artificial neural network (ANN) plays a role of surrogate model of CFD to reduce the computational cost. To improve the performance of ANN, a self-updating logarithm normalized method was proposed to enhance the local prediction of ANN in the inverse design based on GA and ANN. An MD-82 aircraft cabin was used to test the performance of the proposed method, and different environmental parameters were chosen to be the objectives of the cabin environment. The success rate (SR) was used to evaluate the local prediction ability of ANN. Instead of linear normalized ANN, SR was found to be increased by 10.5% with the logarithm normalized ANN and the computational cost was reduced by 23.2% for the same quality of solution.
