Assessing the Performance-Efficiency Trade-off of Foundation Models in Probabilistic Electricity Price Forecasting

2026-04-16 · Source: Machine Learning · Field: Energy & Utilities — Energy Markets & Policy, Renewable Energy Systems · Depth: Expert, quick

Summary

A study compared four models for day-ahead probabilistic electricity price forecasting (PEPF) in European bidding zones: NHITS+QRA, a conditional Normalizing-Flow forecaster (NF), and two Time Series Foundation Models (TSFMs), Moirai and ChronosX. The research found that TSFMs generally outperformed task-specific deep learning models trained from scratch across various market conditions, as measured by CRPS, Energy Score, and predictive interval calibration. However, well-configured task-specific models, specifically NHITS combined with Quantile-Regression Averaging (QRA), achieved performance very close to TSFMs. In certain scenarios, such as when provided with additional informative features or adapted via few-shot learning from other European markets, NHITS+QRA could even surpass TSFMs. The findings highlight a trade-off between the expressive capabilities of TSFMs and the competitive performance of conventional models, emphasizing the need to consider computational expense versus marginal performance gains.

Key takeaway

For data scientists developing electricity price forecasting models, you should carefully evaluate the trade-off between the marginal performance improvements offered by Time Series Foundation Models (TSFMs) and their potentially higher computational expense. Consider starting with well-configured task-specific models like NHITS+QRA, especially when additional features or few-shot learning from other markets can be applied, as they can achieve comparable or even superior accuracy.

Key insights

TSFMs generally outperform task-specific models in PEPF, but well-configured conventional models remain highly competitive.

Principles

• TSFMs offer expressive modeling capabilities.
• Conventional models can achieve competitive performance.
• Consider computational expense vs. marginal gains.

Method

The study compared NHITS+QRA and a conditional Normalizing-Flow forecaster against Moirai and ChronosX TSFMs for day-ahead probabilistic electricity price forecasting in European bidding zones.

In practice

• Use NHITS+QRA for competitive PEPF.
• Incorporate informative feature groups.
• Apply few-shot learning for market adaptation.

Topics

• Probabilistic Electricity Price Forecasting
• Time Series Foundation Models
• NHITS
• Moirai
• ChronosX

Best for: AI Scientist, Research Scientist, Data Scientist

Related on AIssential

• Tyan-WP: A Wind Power Foundation Model for Ultra-Short-Term Probabilistic Forecasting — Tyan-WP is a foundation model for wind power forecasting, achieving accurate zero-shot predictions by integrating static site data and meteorological covariates.
• Electricity price forecasting across Norway's five bidding zones in the post-crisis era — LightGBM excels in Norwegian electricity price forecasting, but external factors are vital for accuracy under market stress.
• Time-Series Foundation Model Embeddings for Remaining Useful Life Estimation — Frozen time-series foundation models like Chronos-2 enable data-efficient RUL estimation with a lightweight regression head.
• Navigating the Safety-Fidelity Trade-off: Massive-Variate Time Series Forecasting for Power Systems via Probabilistic Scenarios — Large-scale power system forecasting requires new benchmarks and models to balance safety and fidelity.
• TTM: Tiny Foundation Models for Multivariate Time-Series Forecasting — Tiny Time Mixers (TTMs) enable efficient, accurate time-series forecasting via compact pre-trained models and multi-level adaptation.
• Interview with AAAI Fellow Yan Liu: machine learning for time series — Physics-informed foundation models enable accurate, adaptable time series analysis, especially in data-scarce scientific and real-world applications.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine Learning.