Conformal prediction for enhanced decision-making of PV Power in electricity markets

This seminar will explore a framework utilizing conformal prediction (CP), an emerging probabilistic forecasting methodology, to assist decision-making for PV power market participants on the day-ahead market (DAM). It will begin by demonstrating how machine learning models can be employed to construct the point predictions based on weather forecasts. Subsequently, various variants of conformal prediction are introduced to quantify the uncertainty of these predictions. The seminar will then explore the application of several market bidding strategies, including Trust-the-forecast, worst-case, Newsvendor and expected utility maximization (EUM), to facilitate decision-making for market participants on the DAM using CP methods. 

Through a case study in the Netherlands, it will highlight how CP when combined with certain bidding strategies can lead to increased profit with minimal energy imbalance, outperforming classical probabilistic forecasting methods, such as linear quantile regression. The seminar will conclude by emphasizing the need to expand the framework to cover other short-term markets, namely intra-day markets.

Coordinators

Olga Kuryatnikova, Ece Karakoyun, Ruben van Beesten