Forecasting Italian daily electricity generation disaggregated by geographical zones and energy sources using coherent forecast combination

TL;DR

This paper introduces a coherent forecast combination method that improves accuracy and consistency in predicting Italian daily electricity generation across different regions and energy sources, addressing hierarchical disaggregation challenges.

Contribution

The paper presents a novel multi-task forecast combination approach that ensures coherence and enhances accuracy in hierarchical energy generation forecasting.

Findings

Outperforms single-expert and traditional reconciliation methods

Ensures coherence across geographical and energy source constraints

Demonstrates superior accuracy on Italian energy data

Abstract

A novel approach is applied for improving forecast accuracy and achieving coherence in forecasting the Italian daily energy generation time series. In hierarchical frameworks such as national energy generation disaggregated by geographical zones and energy sources, independently generated base forecasts often result in inconsistencies across the constraints. We deal with this issue through a coherent balanced multi-task forecast combination approach, which combines unbiased forecasts from multiple experts while ensuring coherence. Applied to the daily Italian electricity generation data, our method shows superior accuracy compared to single-task base and combined forecasts, and a state-of-the-art single-expert reconciliation technique, demonstrating to be an effective approach to forecasting linearly constrained multiple time series.