A Hybrid Strategy for Probabilistic Forecasting and Trading of Aggregated Wind-Solar Power: Design and Analysis in HEFTCom2024

TL;DR

This work presents a practical, reproducible hybrid framework for probabilistic forecasting and day-ahead trading of a wind-solar hybrid, demonstrated in HEFTCom2024. It couples stacked sister NWP forecasts for wind, online post-processing for solar distribution shifts, and a novel aggregated quantile approach to reconstruct total-generation distributions, topped by a stochastic trading strategy and an end-to-end learning option. Case studies show gains in forecast accuracy (via MPL, MCRPS, MWS) and trading revenue, with notable improvements from wind stacking and online solar calibration, and a modest yet meaningful edge from end-to-end trading optimization. The methodologies balance accuracy, computational efficiency, and robustness, offering a practical blueprint for operators seeking probabilistic forecasting and revenue-maximizing decisions in volatile energy markets.

Abstract

Obtaining accurate probabilistic energy forecasts and making effective decisions amid diverse uncertainties are routine challenges in future energy systems. This paper presents the winning solution of team GEB, which ranked 3rd in trading, 4th in forecasting, and 1st among student teams in the IEEE Hybrid Energy Forecasting and Trading Competition 2024 (HEFTCom2024). The solution provides accurate probabilistic forecasts for a wind-solar hybrid system, and achieves substantial trading revenue in the day-ahead electricity market. Key components include: (1) a stacking-based approach combining sister forecasts from various Numerical Weather Predictions (NWPs) to provide wind power forecasts, (2) an online solar post-processing model to address the distribution shift in the online test set caused by increased solar capacity, (3) a probabilistic aggregation method for accurate quantile forecasts of hybrid generation, and (4) a stochastic trading strategy to maximize expected trading revenue considering uncertainties in electricity prices. This paper also explores the potential of end-to-end learning to further enhance the trading revenue by shifting the distribution of forecast errors. Detailed case studies are provided to validate the effectiveness of these proposed methods. Code for all mentioned methods is available for reproduction and further research in both industry and academia.

Figures (10)

• Figure 1: Grid coordinates of historical meteorological data for Hornsea 1 and East England PV plants.
• Figure 2: Overview of the methodology in the forecasting track. It consists of four modules: data pre-processing, wind power forecasting, solar power forecasting, and probabilistic aggregation. The final forecast result is the CDF of total generation, as shown in the curve plot in the lower right corner. The quantile forecasts from 10% to 90% in 10% increments are obtained from the inverse function of the CDF.
• Figure 3: Data pre-processing pipeline for the HEFTCom2024 data.
• Figure 4: The test results of solar power forecasting models trained on historical data. The line “y=x” denotes the point set that the predicted value equals the actual value. The line “LR fit” denotes the linear regression fit of the point set.
• Figure 5: The mean and variability of the price spread at different hour-of-day in various periods. Each subplot represents the statistical data of a half-year period. The x-axis represents different hours of the day. The markers represent the mean of the price spread at each hour, while the bars represent the standard deviation of the price spread at each hour.