1164. Using GANs for the hourly shapping of renewables in Monte-Carlo simulations
Invited abstract in session TB-54: Stochastic Models and Simulation, stream Stochastic modelling.
Tuesday, 10:30-12:00Room: Liberty 1.08
Authors (first author is the speaker)
| 1. | Geoffroy Chaussonnet
|
| Middle Office Trading, EnBW Energie Baden-Württemberg AG | |
| 2. | Andreas Eppler
|
| New Business Pioneer, EnBW Energie Baden-Württemberg AG |
Abstract
In this paper we applied generative adversarial network (GAN) to shape the hourly production of solar and wind power in Germany for mid-term Monte-Carlo simulations.
In Monte-Carlo simulations with a forecast horizon up to the liquid horizon (ca. 3 years), renewable production is usually expressed by the ratio of production normalized by the installed capacity. This gives the advantage of choosing different trajectories for the renewable production market shares.
In a stochastic approach, the daily renewable production ratio can be easily mimicked by simple distributions, which constitute a first step of modeling.
To model the hourly granularity, there are two major caveats.
First, hourly shapes cannot be easily described by simple distributions to match the moments and the autocorrelation. Second, as the hourly shape is also a ratio, multiplying the hourly shape by the daily ratio can lead to production ratio above one.
To solve these two points, a GAN is employed as a shaping engine. It takes renewable daily ratios and behavioral features (dates and holidays) as input and produce multiple hourly shape for each technology. It is shown that this approach generates realistic hourly shapes, in terms distribution, autocorrelation and technology cross correlation.
Keywords
- Stochastic Models
- Electricity Markets
- Simulation
Status: accepted
Back to the list of papers