Improvement of the Training and Normalization Method of Artificial Neural Network in the Prediction of Indoor Environment

Abstract

When designing the indoor environment based on computational fluid dynamics (CFD), the artificial neural network (ANN) playing a role of surrogate model of CFD is involved to reduce the computational cost. To improve the performance of ANN, the training and normalization method of ANN are studied. An MD-82 aircraft cabin is used to test the proposed method, and different environmental parameters are used to evaluate the cabin environment. The results of different training methods are compared, and the parallel combination of genetic algorithm and particle swarm optimization shows better prediction accuracy than other training methods. A local logarithm normalization method is proposed to improve the success rate of ANN prediction. The success rate is increased by 2.5∼11.0% when the proposed local logarithm normalization method is adopted instead of local linear normalization one.