Residential probabilistic load forecasting: A method using Gaussian process designed for electric load data. (15th May 2018)

• Record Type: Journal Article
• Authors: Shepero, Mahmoud; van der Meer, Dennis; Munkhammar, Joakim; Widén, Joakim
• Is Part Of: Applied energy. Volume 218(2018)
• Page Start: 159
• Page End: 172
• Publication Date: 2018-05-15
• Subjects: Gaussian process -- Probabilistic load forecasting -- Residential load; Power (Mechanics) -- Periodicals; Energy conservation -- Periodicals; Energy conversion -- Periodicals; 621.042
• DOI: 10.1016/j.apenergy.2018.02.165
• Languages: English
• ISSNs: 0306-2619

Abstract

Highlights: Probabilistic residential load forecasting using Gaussian and log-normal processes. Deterministic and probabilistic error metrics evaluated the proposed processes. Our results produced sharper forecasts compared with previous models. The log-normal process outperformed the Gaussian process in the forecast sharpness. The log-normal, unlike the Gaussian, process produced a varying forecast sharpness. Abstract: Probabilistic load forecasting (PLF) is of important value to grid operators, retail companies, demand response aggregators, customers, and electricity market bidders. Gaussian processes (GPs) appear to be one of the promising methods for providing probabilistic forecasts. In this paper, the log-normal process (LP) is newly introduced and compared to the conventional GP. The LP is especially designed for positive data like residential load forecasting—little regard was taken to address this issue previously. In this work, probabilisitic and deterministic error metrics were evaluated for the two methods. In addition, several kernels were compared. Each kernel encodes a different relationship between inputs. The results showed that the LP produced sharper forecasts compared with the conventional GP. Both methods produced comparable results to existing PLF methods in the literature. The LP could achieve as good mean absolute error (MAE), root mean square error (RMSE), prediction interval normalized average width (PINAW) and prediction interval coverage probability (PICP) as 2.4%, 4.5%, 13%, 82%, respectively evaluated on the normalized load data.