Abstract: In view of the dominant position of fossil energy in global energy consumption and the environmental problems caused by excessive use of fossil energy, accurate prediction of fossil energy consumption is of great significance for formulating scientific energy policies and optimizing energy structure. Traditional forecasting methods have limitations when dealing with small samples, nonlinear and multi-factor problems, while grey system theory and neural network model are good at dealing with uncertainty and nonlinear mapping respectively. Therefore, this study hybridizes the generalized regression neural network (GRNN) model on the basis of the dynamic nonlinear grey delay multivariable Logistic model, i.e. NGDM(1, N), and constructs grey NGDM(1, N)-GRNN hybrid model to further optimize the prediction results. Particle swarm optimization (PSO) algorithm was used to optimize grey model parameters, and the optimal smoothing factor of GRNN model was found through cross-validation, which improved the prediction accuracy of the model. The empirical results show that compared with the single NGDM(1, N) model and GRNN model, the proposed hybrid model has smaller errors in the short-term prediction of fossil energy consumption, and has better forecasting effect.

Key words: