Impact of BESS last-minutes reactions on short-term system imbalance forecasting accuracy in European energy markets

Abstract

The increasing deployment of Battery Energy Storage Systems (BESS) in modern electricity markets has introduced new complexities in system imbalance (SI) forecasting, particularly due to last-minute balancing actions by Balance Responsible Parties (BRPs). Conventional forecasting models, which primarily rely on historical imbalance patterns and exogenous market features, often fail to capture the dynamic corrective responses of BESS, leading to substantial prediction inaccuracies. This study systematically evaluates the impact of key battery parameters, including maximum power capacity ($P_{max}$), depth of discharge $\left(DOD\right)$, and energy-to-power $(E/P)$ ratio, on forecasting accuracy. A battery-aware autoregressive (AR) model is developed to explicitly integrate these factors, with predictive performance benchmarked against conventional models under both static and dynamic battery dispatch conditions. The analysis establishes well-defined operational stability constraints, demonstrating that forecast errors remain within considerably limits ($\text{MAE}\le 100$ MW) when $P_{max}\approx 416$ MW, $\text{DOD}\le 0.92$, and the $E/P$ ratio is maintained either at $E/P\le 4.64$ or $E/P\ge 6.94$. However, within the intermediate range $4.64<E/P<6.94$, forecast errors exceed 100 MW, introducing instability and reducing predictive reliability. Notably, when $P_{max}$ is below 416 MW, variations in $E/P$ and $DOD$ exhibit minimal influence on forecast accuracy concerning the 100 MW MAE threshold. These findings underscore the intricate interdependencies among BESS parameters, highlighting the destabilizing effects of high-power dispatch, extended storage durations, and deep discharge cycles beyond these defined thresholds. Comparisons against the Elia forecast and a Naïve benchmark confirm that the battery-aware model enhances forecasting accuracy, improving MAE by up to 13.39%, RMSE by 20.61%, and sign accuracy by 12.06% over the Elia baseline. These improvements demonstrate the necessity for forecasting models that explicitly integrate battery dynamics to enhance predictive stability in evolving electricity markets. While this study employs an autoregressive framework for demonstration, the methodology and insights extend to advanced machine learning and probabilistic forecasting approaches. The findings provide actionable guidance for Transmission System Operators (TSOs) and market participants, offering a structured approach for optimizing imbalance management strategies, enhancing grid stability, and improving forecasting resilience in power systems with increasing BESS deployments.