Stacking for Probabilistic Short-term Load Forecasting

TL;DR

This paper investigates stacking point forecasts to produce probabilistic short-term load forecasts (STLF) using three quantile-oriented meta-models: Quantile Estimation through Residual Simulation (QRS), Quantile Linear Regression (QLR), and Quantile Regression Forest (QRF), each with global and local training variants. The authors formalize the problem as learning a function that maps base forecasts to a set of quantiles across a probability grid $\Pi$, and they evaluate the approaches on 35 European countries with triple-seasonality using 16 base models and ENTSO-E data. Empirical results show that QRF delivers the best probabilistic forecasts overall, outperforming QRS and QLR in most cases, with good calibration and competitive point-forecast behavior. The study demonstrates the value of global/local meta-learning in forecast combination and highlights QRF as a strong, robust mechanism for probabilistic STLF, offering practical benefits for system operation and risk management.

Abstract

In this study, we delve into the realm of meta-learning to combine point base forecasts for probabilistic short-term electricity demand forecasting. Our approach encompasses the utilization of quantile linear regression, quantile regression forest, and post-processing techniques involving residual simulation to generate quantile forecasts. Furthermore, we introduce both global and local variants of meta-learning. In the local-learning mode, the meta-model is trained using patterns most similar to the query pattern.Through extensive experimental studies across 35 forecasting scenarios and employing 16 base forecasting models, our findings underscored the superiority of quantile regression forest over its competitors

Figures (4)

• Figure 1: Quality metrics for QRF.
• Figure 2: Comparison of quality metrics for probabilistic STLF.
• Figure 3: Results of the Diebold-Mariano tests for MPQRE.
• Figure 4: Relative frequency.