Predicting Electricity Consumption with Random Walks on Gaussian Processes

TL;DR

The paper tackles short-term electricity consumption forecasting under data scarcity and limited compute by building on Gaussian Processes with a shared mean across time series. It combines MAGMA’s cross-series knowledge with Domino, a frugal random-walk sampling method that navigates sampled GP realizations to reduce training costs while preserving uncertainty estimates. Across synthetic experiments, Domino outperforms MAGMA in MAE and demonstrates potential for scalable, uncertainty-aware forecasting in sparse-data contexts. This approach advances practical probabilistic forecasting for energy systems by reducing computational burden and enabling robust predictions under extreme conditions.

Abstract

We consider time-series forecasting problems where data is scarce, difficult to gather, or induces a prohibitive computational cost. As a first attempt, we focus on short-term electricity consumption in France, which is of strategic importance for energy suppliers and public stakeholders. The complexity of this problem and the many levels of geospatial granularity motivate the use of an ensemble of Gaussian Processes (GPs). Whilst GPs are remarkable predictors, they are computationally expensive to train, which calls for a frugal few-shot learning approach. By taking into account performance on GPs trained on a dataset and designing a random walk on these, we mitigate the training cost of our entire Bayesian decision-making procedure. We introduce our algorithm called \textsc{Domino} (ranDOM walk on gaussIaN prOcesses) and present numerical experiments to support its merits.