The probabilistic forecasting of multivariate electricity price time series requires non-Gaussian forecasting techniques. Generative models based on deep neural networks have been very successfully applied in the machine learning community to model multivariate image distributions. We will explore how these techniques can also be used in the field of electricity price forecasting. Commonalities and differences will be discussed. Two resulting publications, one on probabilistic ensemble post-processing and one on models based on implicit generative copulas, will be presented.
Tim Janke; Florian Steinke: Probabilistic multivariate electricity price forecasting using implicit generative ensemble post-processing. In: 16th International Conference on Probabilistic Methods Applied to Power Systems (PMAPS 2020), (2020). https://doi.org/10.1109/PMAPS47429.2020.9183687
Tim Janke; Mohamed Ghanmi; Florian Steinke: Implicit Generative Copulas. 35th International Conference on Neural Information Processing Systems (NeurIPS 2021), virtual Conference, 07.-10.12.2021, (2021). http://arxiv.org/abs/2109.14567