A Paradigm Shift in Solar Energy Forecasting: A Novel Two-Phase Model for Monthly Residential Consumption

Abstract

Accurate forecasting of residential solar energy consumption is important in electricity production, supply and power dispatch. However, the prediction accuracy of traditional forecasting methods is poor due to the complexity of energy consumption data itself. To remedy this shortcoming, a novel two-stage error correction combined forecasting model is established in this research that is based on traditional linear models, seasonal processing techniques and deep learning models as well as intelligent optimization algorithms by comparing eight combined forecasting models. In addition, this paper also analyzes the combined weight values to further explain the reasons why the proposed combined model outperforms other models. Finally, five benchmark models, three evaluation indicators and one hypothesis testing method are adopted to prove the superiority of the developed model in prediction performance. And empirical results show that the proposed combined model achieves the best accuracy and stability in each aspect of the eight prediction horizons, the MAPE value of 24-step ahead prediction is 2.9053%, which is less than 8.0427%, 6.8818%, 4.2636%, 4.3450% and 4.5901% of SGM (1, 1), SARIMA, Prophet-LSTM, Prophet-RNN and Prophet-BPNN, respectively, indicating the superiority of the developed combined model and adding a better option for energy consumption forecasting in the energy sector.

Keywords: Keywords: Residential solar energy consumption, Multi-step rolling forecasting, Seasonal time series, Combined model.