Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Assessing Electricity Demand Forecasting with Exogenous Data in Time Series Foundation Models

Wei Soon Cheong, Lian Lian Jiang, Jamie Ng Suat Ling

TL;DR

Electricity demand forecasting benefits from exogenous data, but foundation models do not universally outperform traditional baselines across climates and horizons. The authors benchmark multiple cross-channel architectures (MOIRAI, MOMENT, TinyTimeMixers, ChronosX, Chronos-2) against a RevIN-LSTM baseline on Singapore and Australia data at hourly and daily granularities, highlighting strong dependency on architecture and geographic context. Chronos-2 often yields the best zero-shot performance among foundation models, yet in Singapore's stable climate simple baselines can excel for short horizons, while Australia benefits more from exogenous predictors. The study argues for energy-domain foundation models pretrained on diverse electricity contexts and tailored cross-channel mechanisms to reliably harness covariate information in real-world grid forecasting.

Abstract

Time-series foundation models have emerged as a new paradigm for forecasting, yet their ability to effectively leverage exogenous features -- critical for electricity demand forecasting -- remains unclear. This paper empirically evaluates foundation models capable of modeling cross-channel correlations against a baseline LSTM with reversible instance normalization across Singaporean and Australian electricity markets at hourly and daily granularities. We systematically assess MOIRAI, MOMENT, TinyTimeMixers, ChronosX, and Chronos-2 under three feature configurations: all features, selected features, and target-only. Our findings reveal highly variable effectiveness: while Chronos-2 achieves the best performance among foundation models (in zero-shot settings), the simple baseline frequently outperforms all foundation models in Singapore's stable climate, particularly for short-term horizons. Model architecture proves critical, with synergistic architectural implementations (TTM's channel-mixing, Chronos-2's grouped attention) consistently leveraging exogenous features, while other approaches show inconsistent benefits. Geographic context emerges as equally important, with foundation models demonstrating advantages primarily in variable climates. These results challenge assumptions about universal foundation model superiority and highlight the need for domain-specific models, specifically in the energy domain.

Assessing Electricity Demand Forecasting with Exogenous Data in Time Series Foundation Models

TL;DR

Electricity demand forecasting benefits from exogenous data, but foundation models do not universally outperform traditional baselines across climates and horizons. The authors benchmark multiple cross-channel architectures (MOIRAI, MOMENT, TinyTimeMixers, ChronosX, Chronos-2) against a RevIN-LSTM baseline on Singapore and Australia data at hourly and daily granularities, highlighting strong dependency on architecture and geographic context. Chronos-2 often yields the best zero-shot performance among foundation models, yet in Singapore's stable climate simple baselines can excel for short horizons, while Australia benefits more from exogenous predictors. The study argues for energy-domain foundation models pretrained on diverse electricity contexts and tailored cross-channel mechanisms to reliably harness covariate information in real-world grid forecasting.

Abstract

Time-series foundation models have emerged as a new paradigm for forecasting, yet their ability to effectively leverage exogenous features -- critical for electricity demand forecasting -- remains unclear. This paper empirically evaluates foundation models capable of modeling cross-channel correlations against a baseline LSTM with reversible instance normalization across Singaporean and Australian electricity markets at hourly and daily granularities. We systematically assess MOIRAI, MOMENT, TinyTimeMixers, ChronosX, and Chronos-2 under three feature configurations: all features, selected features, and target-only. Our findings reveal highly variable effectiveness: while Chronos-2 achieves the best performance among foundation models (in zero-shot settings), the simple baseline frequently outperforms all foundation models in Singapore's stable climate, particularly for short-term horizons. Model architecture proves critical, with synergistic architectural implementations (TTM's channel-mixing, Chronos-2's grouped attention) consistently leveraging exogenous features, while other approaches show inconsistent benefits. Geographic context emerges as equally important, with foundation models demonstrating advantages primarily in variable climates. These results challenge assumptions about universal foundation model superiority and highlight the need for domain-specific models, specifically in the energy domain.
Paper Structure (36 sections, 1 figure, 4 tables)

This paper contains 36 sections, 1 figure, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Local vs Global training paradigm
  4. Local training.
  5. Global training.
  6. Modeling cross-channel relationships
  7. MOIRAI.
  8. TinyTimeMixers (TTM).
  9. MOMENT.
  10. ChronosX.
  11. Chronos-2.
  12. Methodology
  13. Data Collection and Preprocessing
  14. Feature selection.
  15. Experimental configuration
  16. ...and 21 more sections

Figures (1)

  • Figure 1: Top 12 absolute feature correlations by dataset. Selected features are colored in blue.