Energy load forecasting using Terna public data: a free lunch multi-task combination approach

TL;DR

This paper introduces a multi-task stacking regression method that significantly improves 15-minute energy load forecasts for Italy by combining Terna's public data with naive forecasts, enhancing accuracy immediately after data release.

Contribution

The paper presents a novel multi-task stacking regression approach that leverages public energy data and naive forecasts to improve short-term load prediction accuracy.

Findings

Forecast accuracy improves immediately after data publication.

Multi-task stacking outperforms individual naive forecasts.

Method is simple and effective for real-time energy load prediction.

Abstract

We propose a quick-and-simple procedure to augment the accuracy of 15-minutes Italian load forecasts disaggregated by bidding zones published by Terna, the operator of the Italian electricity system. We show that a stacked-regression multi-task combination approach using Terna and daily random walk naive forecasts, is able to produce significantly more accurate forecasts immediately after Terna publishes on its data portal the energy load measurements for the previous day, and the forecasts for the current day.