AI is transforming electricity price forecasting by enabling models to process vast market and operational data, improving accuracy across multiple markets and time horizons.

Forecasting approaches have evolved from statistical models to machine learning, deep learning, and foundation models, with transformer-based architectures excelling at long-range dependencies and multi-horizon outputs.

Foundation models, including PriceFM trained on data from 24 European countries, 38 bidding zones, and four years of data, offer probabilistic price forecasts and capture cross-border interdependencies via a Graph Neural Network component.

Hybrid and ensemble methods blend the stability and interpretability of traditional models with the flexibility of deep learning, often leading in day-ahead and real-time forecasts.

Key inputs include historical market prices and volumes, grid conditions, weather and renewable forecasts, market fundamentals, outages, and regulatory variables.

Applications span day-ahead, intraday, and balancing markets (FCR, aFRR, mFRR), enabling cross-market flexibility optimization and more informed decision-making.

Uncertainty quantification is central in volatile markets, with probabilistic outputs such as ensemble quantiles, distribution forecasts, and conformal prediction providing better decision support for operators and traders.

Author: Peter Nemček, CTO and co-founder of CyberGrid.

The future benefits include improved operational efficiency, grid resilience, and optimal use of flexible assets like batteries, as AI-guided insights help market participants place resources in the right market at the right time.