In the context of increasingly tight energy supply and rising prices, it is of great significance to carry out research on energy consumption prediction models with energy conservation as the goal. In order to improve energy efficiency, it is not only necessary to conduct statistics and analysis on energy historical data, but also to predict future energy data. In this paper, Bicubic interpolation algorithm and convolutional neural network are used to spatially predict energy consumption. The model framework structure for energy consumption prediction is given, and the two models are experimentally analyzed. The results show that the convolutional neural network is better than the Bicubic interpolation method, and the prediction result is closer to the actual value. For the high spatial complexity of energy consumption data, the definition and framework of the RessNetGAN model are first given. Secondly, the specific structure of the generator is given. How to extract the spatial eigenvalue of energy consumption through 3D convolution operation is introduced, and the experiment is carried out by changing the pooling parameters of the energy consumption to be predicted. The results show that the RessNetGAN model with generated confrontation structure has better prediction performance than the convolutional neural network model, and when the input energy consumption data is very low precision, there will be no serious distortion of the prediction result. INDEX TERMS Energy consumption forecasting, deep learning, generating confrontation networks, spatial domain reconstruction.