Renewable Energy Scenario Generation Using Conditional GAN with Multivariate Meteorological Data

The intermittency of renewable energy challenges power system stability. This paper proposes a Conditional Generative Adversarial Network (CGAN) method using multivariate meteorological data for scenario generation. Enhanced by Wasserstein GAN with Gradient Penalty (WGAN-GP), it ensures stable training and high-quality outputs. Using the Gefcom2014 dataset, the model generates photovoltaic day-ahead scenarios with 95.83% coverage and 0.82 kW power interval width, outperforming Markov chain, Monte Carlo, and VAE models. Autocorrelation confirms temporal consistency. This approach reduces historical data reliance, supporting power system planning and showcasing deep learning's potential in renewable energy.