Short-term power load probabilistic interval multi-step forecasting based on ForecastNet. (August 2022)

• Record Type: Journal Article
• Authors: Li, Yupeng; Guo, Xifeng; Gao, Ye; Yuan, Baolong; Wang, Shoujin
• Is Part Of: Energy reports. Volume 8(2022)Supplement 5
• Page Start: 133
• Page End: 140
• Publication Date: 2022-08
• Subjects: Short-term load forecasting -- ForecastNet -- Gaussian process -- Probabilistic forecasting; Power resources -- Periodicals; Energy industries -- Periodicals; Power resources; Periodicals; Electronic journals; 621.04205
• DOI: 10.1016/j.egyr.2022.02.159
• Languages: English
• ISSNs: 2352-4847

Abstract

Abstract: Many uncertain factors to the planning and distribution of the power grid have been brought by connecting to the distributed power grid and increasing active loads. To obtain more accurate and comprehensive information of power load forecasting value, a short-term power load-interval multi-step forecasting method based on ForecastNet is proposed in this paper. Firstly, single variable historical load data is used as input. Secondly, ForecastNet's deep feedforward architecture is proposed to exactly capture the time-varying characteristics of load. Finally, the Gaussian distribution output is used to realize the uncertainty evaluation of the model. Deterministic point forecasting combines probabilistic forecasting to quantify the uncertainty of forecast results. Output power load forecasts in the form of probability intervals. The experimental results show that the ForecastNet forecasting model proposed in this paper outperforms the other three models in the four seasons. In the autumn when the load curve volatility and random components are large, MASE=0.357, SMAPE=4.869%, NRMSE=11.804%. It has the advantages of the high predictive range quality, high predictive accuracy, and great practical engineering value.