Model analysis of energy consumption data for green building using deep learning neural network

Abstract The purposes are to solve the defects of traditional backpropagation neural network (BPNN), such as inclined local extremum and slow convergence, as well as the incomplete data acquisition of building energy consumption (EC). Firstly, a green building (GB)-oriented EC data generation model based on generative adversarial networks (GANs) is implemented; GAN can learn the hidden laws of raw data and produce enhanced virtual data. Secondly, the GB-oriented EC prediction model based on Levenberg Marquardt-optimized BPNN is implemented and used for building EC prediction. Finally, the effectiveness of the proposed model is verified by real building EC data. The results show that the data enhanced by the GAN model can reduce the model prediction error; the optimized BPNN model has lower prediction error and better performance than other models. The purpose of this study is to provide important technical support for the improvement and prediction of GB energy data.