Student: Robin Köster
Supervisor: Tim Janke
Time period: 08/01/2018 - 10/31/2018
Type: Project Seminars Bachelor
Generative Adversarial Networks (GAN) are among the most discussed topics in machine learning right now. Instead of minimizing an explicit loss function, a generative model, the generator, is trained to “fool” another model, the discriminator, by generating realistic samples from a random input vector. On the other hand, the discriminator is trained to distinguish fake samples created by the generator from real samples. Building on this, Conditional GANs generate samples conditionally on an input vector and therefore can be used to make predictions, e.g. on video frames (Lottner, Kreiman, Cox, 2015). This thesis aims to explore the possibilities GANs and C-GANs offer to scenario generation and forecasting for electricity markets, especially intraday trading.