The Hybrid Renewable Energy Forecasting and Trading Competition 2024 Journal: arXiv

Jethro Browell, Dennis van der Meer, Henrik Kälvegren, Sebastian Haglund, Edoardo Simioni, Ricardo J. Bessa, Yi Wang · 2025

Read The Hybrid Renewable Energy Forecasting and Trading Competition 2024 Journal: arXiv by Jethro Browell, Dennis van der Meer, Henrik Kälvegren, Sebastian Haglund, Edoardo Simioni, Ricardo J. Bessa, Yi Wang (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2507-01579.txt.

Introduction

Jethro Browell and co-authors report a 2024 competition that asked teams to forecast and trade wind and solar power in Britain, in a 2025 paper. Teams handled a portfolio of farms rated at 3.6 gigawatts over three months. Each day they filed real forecasts and market bids for the day ahead, mimicking live operations.

Prizes covered forecast accuracy, trading revenue, and combined rank. Popular tree-boosting methods again proved hard to beat for day-ahead forecasts, though other approaches scored well and details of setup mattered most. Trading results showed forecast skill alone does not set earnings; strategy counts too. All data, from weather to bids, are shared openly for benchmarking and further study.

Catalog record

CREATORJethro Browell, Dennis van der Meer, Henrik Kälvegren, Sebastian Haglund, Edoardo Simioni, Ricardo J. Bessa, Yi Wang
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE54 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2507-01579

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

The Hybrid Renewable Energy Forecasting and Trading Competition 2024 Journal: arXiv Jethro Browell, Dennis van der Meer, Henrik Kälvegren, Sebastian Haglund, Edoardo Simioni, Ricardo J. Bessa, Yi Wang Abstract The Hybrid Energy Forecasting and Trading Competition challenged participants to forecast and trade the electricity generation from a 3.6GW portfolio of wind and solar farms in Great Britain for three months in 2024. The competition mimicked operational practice with participants required to submit genuine forecasts and market bids for the day-ahead on a daily basis. Prizes were awarded for forecasting performance measured by Pinball Score, trading performance measured by total revenue, and combined performance based on rank in the other two tracks. Here we present an analysis of the participants’ performance and the learnings from the competition. The forecasting track reaffirms the competitiveness of popular gradient boosted tree algorithms for day-ahead wind and solar power forecasting, though other methods also yielded strong results, with performance in all cases highly dependent on implementation. The trading track offers insight into the relationship between forecast skill and value, with trading strategy and underlying forecasts influencing performance. All competition data, including power production, weather forecasts, electricity market data, and participants’ submissions are shared for further analysis and benchmarking. 1 Introduction Forecasting production from wind and solar power plants, and making effective decisions under forecast uncertainty, are essential capabilities in low-carbon energy systems. Power system operators and energy traders typically rely on a combination of in-house forecasting systems and external forecasting services. Decisions informed by forecast information are generally taken by humans but are increasingly supported or automated by software. These topics are the subject of much academic research and commercial innovation, and while both continuously report performance improvement, it is difficult to know how different approaches compare in practice. Different datasets, evaluation criteria, and the possibility of accidental (or deliberate) data leakage and misreporting all make comparisons challenging ( Möhrlen et al. 2023 ) . This problem is common across many domains and has motivated forecasting competitions to establish best practices by providing a common task and evaluation criteria, with competitions hosted by third parties to ensure accuracy and fairness ( Hyndman 2020 ; Hong et al. 2020 ) . In the energy domain, the Global Energy Forecasting competitions of 2012, 2014 and 2017 have been particularly influential in establishing methodologies for wind, solar, price and load forecasting and for stimulating interest in probabilistic forecasting ( Hong et al. 2014 ; Hong et al. 2016 ; Hong et al. 2019 ) . Other energy forecasting competitions include those run by the European Energy Market conference, including EEM 2020, focused on national wind power production ( Bellinguer et al. 2020 ; Browell et al. 2020 ) , which has received relatively little attention in academic literature; the impact of COVID-19 lockdowns on electricity demand motivated ( Farrokhabadi et al. 2022 ) , promoting the need for methods that can adapt to abrupt changes in underlying behaviours; the BigDEAL Challenge 2022 focused on short-term peak load forecasting, emphasizing the timing and shape of daily peaks, and introduced novel error metrics to benchmark models under realistic operational conditions ( Shukla and Hong 2024 ) ; and two competitions on smart meter forecasting in 2020 and 2021 ( Pekaslan et al. 2023 ) were motivated by challenges related to billing in electricity retail. These more recent examples targeted specific challenges in energy forecasting and go beyond the mature practice of day-ahead forecasting for individual wind or solar plants, or forecasting total national/system load ( Hong et al. 2020 ) . The Hybrid Renewable Energy Forecasting and Trading Competition (HEFTcom) was motivated by the learning and community benefits that previous competitions produced, as well as the potential for new competition to contribute to open research questions. Specifically, the design of HEFTcom was guided by the following key objectives: 1. Encourage novel forecasting models tailored to combined wind and solar energy portfolios in an operational setting. 2. Evaluate recent forecasting advances, including deep learning, versus standard methods in an operational setting for renewable energy trading. 3. Study how forecasting accuracy impacts its value in a decision-making problem under uncertainty. 4. Assess the complexity of the energy trading decision chain, focusing on the number of forecasting and bidding models involved in generating the optimal bid. 5. Produce an open dataset for benchmarking. First, we aim to stimulate the development of novel forecasting methods for hybrid portfolios of wind and solar power, and establish best practices for this task. Hybrid generation forecasting represents a novel aspect compared to past competitions, such as GEFcom, which focused on single technologies. As demonstrated in ( Couto and Estanqueiro 2023 ) , it poses unique challenges, such as identifying the most relevant weather parameters. Furthermore, previous competitions, notably GEFcom2014, popularised tree-based methods for wind and solar power forecasting but also highlighted the importance of forecaster expertise, data quality, preprocessing strategies, and validation techniques employed by each team ( Hong et al. 2016 ) . While tree-based methods have consistently performed well in structured tabular data, including in the M5 forecasting competition ( Makridakis et al. 2022 ; Januschowski et al. 2022 ) , recent advances in deep learning and other machine learning algorithms, especially those tailored for time series forecasting, require ongoing empirical comparison. However, the aim is not to claim superiority of one class of models over another, but to encourage researchers to systematically benchmark emerging techniques against established ones. This helps uncover potential innovations while accounting for the significant influence of forecaster expertise, data preparation, and evaluation methodology on performance. Secondly, the use of forecasts in decision-making and the connection between forecast performance and value are poorly understood and warrant attention. A forecast (or forecast ‘improvement’) only has value if it leads to better decisions ( Möhrlen et al. 2023 ; Pinson et al. 2007 ) . HEFTcom aimed to advance this discussion by integrating both predictive modelling and downstream application into its structure. Energy trading was a natural choice for the decision-making problem as it has an inherent scoring mechanism, market revenue, and provides a link to other energy forecasting problems including price and volume forecasting. Notably, the M6 forecasting competition in financial forecasting introduced a novel evaluation approach by assessing both probabilistic forecast accuracy (e.g., Ranked Probability Score) and the effectiveness of forecasts in portfolio optimization using metrics like the Information Ratio ( Makridakis et al. 2024a ; Makridakis et al. 2024b ) . Inspired by such initiatives, HEFTcom encourages participants to consider not only how well forecasts perform statistically but also how they influence the decision-making outcomes. A further objective was to assess the complexity of the decision-making model chain, particularly the number and type of models combined to produce the submitted bid. This includes cases where multiple power and price forecasting models are integrated with decision models (e.g., stochastic optimization, heuristics rules), as well as more prescriptive approaches where a single model directly prescribes the optimal bid ( Carriere and Kariniotakis 2019 ) . Through this, HEFTcom seeks to understand on how forecasters bridge the gap between predictive analytics and actionable decisions, a topic with growing relevance in energy systems research. The final design consideration was the practical applicability of solutions.

Read the complete text