Forecasting the residential solar energy consumption of the United States

In recent years, residential solar energy consumption of United States under the effect of a series of encouragement policies has exhibited a growth trend characterized by seasonal leap. To predict it, a new grey model based on data grouping and buffer operator is proposed. The model groups on a monthly or quarterly basis are grouped, which are then buffered separately to cope with prediction error caused by seasonal fluctuations and sudden changes in trend. In addition, a genetic algorithm is used to obtain the most appropriate degree of buffering. And then, the predictive effects of classical grey model, grey model based on data grouping, non-linear autoregressive neural network, echo state network, and the proposed model are compared. The results show that the mean absolute percentage errors of predicted results obtained by using these five models are 32.73%, 30.23%, 46.94%, 39.15%, and 6.17%, respectively, implying that the proposed model confers a significant advantage. Compared with the other four models, the new model can more effectively recognize the seasonal fluctuation and structural mutation of time series data. After conducting out-of-sample forecasting, the results demonstrate that the residential solar energy consumption of United States will maintain its rapid growth with an average annual growth rate of 24%.