学术文献库

Power producers use a wide range of decision support systems to manage and plan for sales in the day-ahead electricity market, and they are often faced with the challenge of choosing the most advantageous bidding strategy for any given day. The optimal solution is not known until after spot clearing. Results from the models and strategy used, and their impact on profitability, can either continuously be registered, or simulated with use of historic data. Access to an increasing amount of data opens for the application of machine learning models to predict the best combination of models and strategy for any given day. In this article, historical performance of two given bidding strategies over several years have been analyzed with a combination of domain knowledge and machine learning techniques (gradient boosting and neural networks). A wide range of variables accessible to the models prior to bidding have been evaluated to predict the optimal strategy for a given day. Results indicate that a machine learning model can learn to slightly outperform a static strategy where one bidding method is chosen based on overall historic performance.