Probabilistic energy forecasting through quantile regression in reproducing kernel Hilbert spaces

Luca Pernigo (Università della Svizzera italiana); Rohan Sen (Università della Svizzera italiana); Davide Baroli (Università della Svizzera italiana); Davide Baroli (Università della Svizzera italiana)

Abstract

To ensure sustainable and resilient energy development, accurate forecasting of energy demand is essential. Achieving the Net Zero RCP 4.5 intermediate pathway in the DACH countries requires an increase in renewable energy production, energy storage and a reduction in building consumption. The realisation of this scenario depends on hydropower capacity and climate variables. To make an informed decision, we need to quantify the uncertainty in the intermittency of energy demand, which has led to an increase in research on probabilistic forecasting (PF) in the short and medium horizon.
In this study, we explore a non-parametric method based on the framework of reproducing kernel Hilbert spaces (RKHS), known as the kernel quantile regression, for energy PF. Our experiments demonstrate the reliability and sharpness of this method, and we benchmark it against state-of-the-art methods in the literature. Our numerical experiments indicate that this method produces superior accuracy in predicting load and price. In the numerical benchmarks, we extend the GEFCOM2014 benchmark for DACH counties, including the hydro-power explanatory variable, and we forecast load using the recent SECURES-Met dataset. We provide our implementation along with complementary scripts for reproducible research.