Conditional Generative Adversarial Network-based framework for multi-feature uncertainty modeling in energy systems
article
This paper presents a conditional Generative Adversarial Network (cGAN)-based framework for capturing uncertainty in the portfolio management of a hybrid power plant, with a particular focus on the joint variability of wind power output and electricity market prices. The proposed cGAN model generates realistic scenarios for multiple correlated feature vectors simultaneously, while preserving both temporal dependencies and inter feature correlations. A large set of scenarios is produced and subsequently reduced to a limited number of representative scenarios using a clustering technique that retains the statistical structure and correlations among variables. These representative scenarios are then integrated into a developed stochastic Mixed-Integer Linear Programming (MILP) model within the EMERGE platform at TNO to optimize hybrid power plant portfolio management under uncertainty. Results based on multi-year data demonstrate that the approach reduces imbalance costs from 20.63% to 14.94% compared to a deterministic baseline that relies only on point forecasts, which highlights the effectiveness of the proposed framework in enhancing operational robustness and market alignment.
Topics
TNO Identifier
1018817
Source
Electric Power Systems Research(251), pp. 1-8.
Pages
1-8