Abstract
Inspired from recent insights into the common ground of machine learning,optimization and decision-making, this paper proposes an easy-to-implement, buteffective procedure to enhance both the quality of renewable energy forecastsand the competitive edge of renewable energy producers in electricity marketswith a dual-price settlement of imbalances. The quality and economic gainsbrought by the proposed procedure essentially stem from the utilization ofvaluable predictors (also known as features) in a data-driven newsvendor modelthat renders a computationally inexpensive linear program. We illustrate theproposed procedure and numerically assess its benefits on a realistic casestudy that considers the aggregate wind power production in the Danish DK1bidding zone as the variable to be predicted and traded. Within this context,our procedure leverages, among others, spatial information in the form of windpower forecasts issued by transmission system operators (TSO) in surroundingbidding zones and publicly available in online platforms. We show that ourmethod is able to improve the quality of the wind power forecast issued by theDanish TSO by several percentage points (when measured in terms of the meanabsolute or the root mean square error) and to significantly reduce thebalancing costs incurred by the wind power producer.