Deep Learning for Cross-Border Electricity Price Forecasting: A Comparative Study

Hadeer Elashhab; Sai Srijan Papineni; Marvin Dorn; Veit Hagenmeyer; Benjamin Schäfer

cs.LGAug 17, 2026

Deep Learning for Cross-Border Electricity Price Forecasting: A Comparative Study

Authors: Hadeer Elashhab, Sai Srijan Papineni, Marvin Dorn, Veit Hagenmeyer, Benjamin Schäfer

Organizations: Institute for Automation and Applied Informatics, Karlsruhe Institute of Technology, Karlsruhe, Germany · Department of Computer Science, Karlsruhe Institute of Technology, Karlsruhe, Germany

Abstract

While publicly available electricity market data presents a valuable resource for forecasting research, the field lacks established benchmark datasets for standardized comparison. As a result, many studies have relied on different datasets and metrics to evaluate methods in isolated settings, making it difficult to assess progress and compare state-of-the-art approaches consistently. In this work, we use public data to evaluate deep learning models for electricity price forecasting (EPF) across multiple market settings. Our goal is to establish a reproducible framework that enables a consistent evaluation of forecasting models. Although deep learning has been explored for day-ahead EPF, many prior studies are limited to single-market settings, narrow feature sets, or fixed training regimes. This work presents a comparative evaluation of six deep learning models--covering state-space, MLP, RNN, and Transformer-based architectures--emphasizing generalization across markets. We simulate low-data target-market conditions using zero-shot, one-shot, and few-shot learning. Our test set focuses on the Germany-Luxembourg (DE-LU) bidding zone in 2024 using a standardized dataset with calendar, historical price, and market-derived features. Our findings suggest that N-HiTS and NBEATSx perform competitively in limited-data scenarios, while transformer-based models can reach comparable accuracy but tend to require more adaptation and tuning. Model performance also benefits from careful feature selection and hyperparameter tuning, and we note that the differences between the strongest models are often small.

Figures & tables

Fig. 1 : Average electricity spot market prices across European bidding zones in 2024 (EUR/MWh). The color indicates average annual price levels, with dark red representing higher prices and dark green representing lower prices. Bidding zones labeled with their respective codes (e.g., DE-LU, FR, DK1) are those included in the learning strategy experiments presented in this work. Adapted from [ 22 ] with permission.

Fig. 2 : Day-ahead electricity prices (EUR/MWh) for the DE-LU bidding zone. The training data (blue) spans from 2018-10-01 to 2023-12-31, while the test data (red) contains prices for 2024. The figure highlights extreme price spikes, particularly during the 2022 energy crisis, where surging gas prices had a direct impact on electricity costs. Notably, the highest recorded prices are observed in the test dataset, reflecting the day-ahead electricity spot market prices of 2024.

Learning Strategy	DE-LU Training Data	Purpose
Zero-Shot (ZSL)	0 days	Full generalization
One-Shot (OSL)	7 days + 1 forecast window	Minimal adaptation
Few-Shot (FSL)	4 weeks	Partial adaptation

TABLE I : Overview of DE-LU exposure under each learning strategy.

Fig. 3 : Comparison of actual prices (EUR/MWh) vs. the Synthetic Price baseline, which is derived from gas prices and CO 2 emission allowances.

Baseline Model	MAE	RMSE
Previous Day’s Price	27.85	44.35
Synthetic Price	31.97	51.49
Previous Week’s Price	32.40	55.11
Previous Month’s Price	33.01	51.22

TABLE II : Evaluation of Simple Baseline Models using MAE and RMSE. The best-performing model for each metric is highlighted in blue .

Panel of Figure 6 — Fig. 4 : Comparison of different baseline models (blue) against actual electricity prices (orange) for 2024 in EUR/MWh. The Previous Day’s Price performs best in stable periods.

Model	Start year	MAE	MAE	RMSE	RMSE	Input days
		(EP)	(EP+Cal)	(EP)	(EP+Cal)
Mamba	2018	27.45	22.72	44.36	39.31	7
NBEATSx	2023	22.32	22.10	37.34	37.48	7
N-HiTS	2020	23.8	21.9	45.02	37.42	7
LSTM	2023	22.92	24.88	38.77	43.70	7
TFT	2018	23.14	21.69	40.22	36.13	7

TABLE III : Model performance with and without calendar features (EP = electricity price only; Cal = calendar features). Best value per column is bold.

Model	Year	MAE	RMSE	Gas	Synth	CO 2	Load
Mamba
MAE: 22.72	2018	22.51	38.89	✓
RMSE: 39.31	2018	26.84	44.94	✓	✓
NBEATSx
MAE: 22.10	2023	20.90	35.81	✓
RMSE: 37.34	2023	20.99	36.27	✓	✓

TABLE IV : Mini ablation study of exogenous features by model. Each row shows a different input combination. Best MAE and RMSE for each model are shown in bold. Under each model’s name is its previous best MAE and RMSE value.

Model	Year	MAE	RMSE	SP features	Load
NBEATSx	2020	19.11	34.93	✓
NBEATSx	2020	20.09	35.28	✓	✓
N-HiTS	2020	18.75	34.77	✓
N-HiTS	2020	21.70	37.29	✓	✓
TFT	2018	19.40	35.63	✓
TFT	2018	20.51	35.88	✓	✓

TABLE V : Model performance with and without SP features and load during hyperparameter optimization. SP features include gas price, CO 2 emission allowances, and synthetic price. Models without these use only electricity price and calendar features. Bold values indicate the lowest value per method if there are multiple values.

Fig. 5 : Illustration of a forecast using the NBEATSx model with One-Shot Learning (OSL). Achieving an MAE of 18.48 EUR/MWh, this model closely tracks price dynamics with only few exceptions. The orange line represents the actual electricity price, while the blue line shows the NBEATSx’s prediction.

Model	Zero-shot	One-shot	Few-shot
N-HiTS	18.61	18.65	18.61
NBEATSx	18.96	18.48	18.57
TFT	18.63	19.35	19.13
VT	18.50	18.77	18.96

TABLE VI : MAE results for zero-shot, one-shot, and few-shot learning across models. Bold values here indicate lowest value per strategy or column.

Hadeer Elashhab received her B.S. degree in computer science from the American University in Cairo, Cairo, Egypt, in 2017 and her M.S. degree in artificial intelligence from Queen Mary University of London, London, United Kingdom, in 2022. She is currently pursuing her Ph.D. degree in computer science at Karlsruhe Institute of Technology, Karlsruhe, Germany. From 2017 to 2020, she was a Data Scientist with Raisa Energy LLC, Cairo, Egypt. Her research interest includes the application of machine learning and deep learning techniques to energy time series, explanation of applied machine and deep learning methods, and the evaluation of explanation quality produced by such methods.

Table 13

Sai Srijan Papineni received the B.Sc. degree in computer science from the Karlsruhe Institute of Technology (KIT), Karlsruhe, Germany, in 2025. He is currently a Developer at SAP SE, Walldorf, Germany. Previously, he worked as a Working Student at SAP SE during his undergraduate studies. Mr. Papineni’s research interests include deep learning–based time series forecasting, electricity price modeling, and the application of machine learning methods to energy markets.

Table 14

Veit Hagenmeyer Veit Hagenmeyer received a diploma in technical kybernetics from the University of Stuttgart, Stuttgart, Germany, in 1998, and a Docteur en Science in engineering from Université Paris, Paris, France, in 2002. His primary research interests are modeling, optimization, and control of sector-integrated energy systems, machine-learning-based forecasting for renewable-driven energy networks, and integrated cyber-security of such systems. He is currently Professor of Energy Informatics with the Faculty of Informatics and Director of the Institute for Automation and Applied Informatics at the Karlsruhe Institute of Technology (KIT), Karlsruhe, Germany. From 2010 to 2014, he led three BASF power-plant sites and their grids (approximately 1 GW, about 200 staff, and a 750 million Euros annual budget). Earlier he served as Personal Assistant to the European Location Manager of BASF (2009–2010), Leader of the Advanced Process Control Group at BASF (2007–2008), Research Engineer in the same group (2003–2006), researcher on nonlinear control theory at the Institute of System Dynamics, University of Stuttgart (2003), and Postdoctoral Fellow on nonlinear control theory at L2S, C.N.R.S.-Supélec-Université, Paris (2002). His current research continues to focus on integrated energy-system modeling, data-driven forecasting, and cyber-security. Prof. Hagenmeyer is a member of the ACM e-Energy steering committee (2025) and holds several recognitions, including the ACM e-Energy Most Accepted Author award (2024), Best Reviewer award (2023), multiple KIT Faculty Awards for best teaching (2021, 2020, 2019, 2018, 2016), the BASF Innovation Awards Finalist (2009), the Lindau Meetings of Nobel Laureates Participation Award (2006), and the VDI Eugen-Hartmann Award (2005).

Table 17

Appendix figures & tables1 asset

Supplementary material from the paper’s appendix.

Appendix

Model	Training Time (minutes)
Mamba	10–25
LSTM	15–30
N-HiTS, NBEATSx	20–40
VT, TFT	40–50

Appendix

TABLE VII : Comparison of model training times. Times are provided as approximate ranges and measured on a GPU-enabled system.

Explore similar work

Foundation models for electricity price forecasting and battery arbitrage: Can they replace market-specific forecasting models?
Aug 31, 2026Arkadiusz Lipiecki, Rafał WeronTabular Foundation Models Electricity Price Forecasting
Department of Computational Social Science, Wrocław University of Science and Technology, Poland · Department of Operations Research and Business Intelligence, Wrocław University of Science and Technology, Poland · Center for Research in Energy (CoRE), Aarhus University, 8000 Aarhus C, Denmark
An Empirical Benchmark of Deep Time-Series Models for Smart Meter Energy Forecasting
Aug 19, 2026Behnaz Kavoosighafi, Maria Eidenskog, Wiktoria Glad +1Time Series Forecasting Forecasting Benchmarks
Linköping University, Department of Science and Technology, Norrköping, 60174, Östergötland, Sweden · Linköping University, Department of Thematic Studies, Linköping, 58183, Östergötland, Sweden
Empirical evaluation of Time Series Foundation Models for Day-ahead and Imbalance Electricity Price Forecasting in Belgium
May 16, 2026Chi Bui, Maria Margarida Mascarenhas, Arnaud Verstraeten +1Zero-Shot Time Series Forecasting Electricity Price Forecasting
ELECTA-ESAT (KU Leuven) Leuven, Belgium · ELECTA-ESAT(KU Leuven) and Gridual Leuven, Belgium

Aug 31, 2026J/K move · Enter open · S save

Foundation models for electricity price forecasting and battery arbitrage: Can they replace market-specific forecasting models?

Arkadiusz Lipiecki, Rafał Weron

Department of Computational Social Science, Wrocław University of Science and Technology, Poland · Department of Operations Research and Business Intelligence, Wrocław University of Science and Technology, Poland · Center for Research in Energy (CoRE), Aarhus University, 8000 Aarhus C, Denmark

Tabular Foundation Models Electricity Price Forecasting Time Series Forecasting Probabilistic Forecasting Time-Series Foundation Models

Open paper AskPDF arXiv

Foundation models promise accurate forecasts with little or no task-specific training, but whether they can replace models designed specifically for electricity price forecasting remains unclear. We compare nine variants from five foundation model families, evaluated in zero-shot mode, with two state-of-the-art electricity price forecasting benchmarks in Germany, Poland, and Spain over 2021-2025. Their performance is assessed in terms of point and probabilistic forecasting accuracy, as well as economic value in battery energy storage arbitrage. Only the TabPFN models consistently and significantly outperform the benchmarks across all three markets and all statistical measures. However, this statistical dominance does not translate directly into economic dominance: TabPFN performs best under unlimited bids and riskier quantile-based strategies, whereas the Distributional Deep Neural Network benchmark is more profitable when risk tolerance is lower. Thus, foundation models cannot universally replace market-specific models, and their value depends on both model architecture and the decision problem.

Abstract

Explore similar work

Foundation models for electricity price forecasting and battery arbitrage: Can they replace market-specific forecasting models?

An Empirical Benchmark of Deep Time-Series Models for Smart Meter Energy Forecasting

Empirical evaluation of Time Series Foundation Models for Day-ahead and Imbalance Electricity Price Forecasting in Belgium