Researchers have developed and evaluated eight different forecasting models, including LightGBM and deep learning architectures, to predict electricity prices across Norway's five bidding zones. The study utilized a multimodal hourly dataset from 2019 to 2025, employing rigorous backtesting and feature ablation techniques. LightGBM emerged as the top-performing model in all zones, with simpler models relying on lagged prices and calendar variables also showing strong accuracy. However, external factors like reservoir levels and gas prices were found to be essential for refining forecast accuracy, particularly during market stress. AI
Summary written by gemini-2.5-flash-lite from 3 sources. How we write summaries →
IMPACT Provides benchmarked forecasting models for energy markets, highlighting the importance of external features during market stress.
RANK_REASON Academic paper on forecasting models with benchmark results.