Automated photovoltaic forecasts with limited information using an ensemble of pre-trained models.
AutoPV addresses three challenges:
The underlying idea of AutoPV is to describe the arbitrary mounting configuration of a new PV plant as a convex linear combination of outputs from a sufficiently diverse ensemble pool of PV models of the same region. AutoPV incorporates three steps: i) create the ensemble model pool, ii) form the ensemble output by an optimally weighted sum of the scaled model outputs in the pool, and iii) rescale the ensemble output with the new PV plant’s peak power rating.
To install this project, perform the following steps.
Exemplary evaluations using AutoPV are given in the examples folder.
If you use this method please cite the corresponding paper:
Stefan Meisenbacher, Benedikt Heidrich, Tim Martin, Ralf Mikut, and Veit Hagenmeyer. 2023. AutoPV: Automated photovoltaic forecasts with limited information using an ensemble of pre-trained models. In Proceedings of the Fourteenth ACM International Conference on Future Energy Systems (e-Energy ’23). Association for Computing Machinery, New York, NY, USA, 386–414. https://doi.org/10.1145/3575813.3597348
This project is funded by the Helmholtz Association under the Program “Energy System Design” and the Helmholtz Association’s Initiative and Networking Fund through Helmholtz AI.
The example data includes weather measurements from DWD:
Deutscher Wetterdienst. 2023. Historical 10-minute station observations of solar incoming radiation, longwave downward radiation, pressure, air temperature, and mean wind speed for Germany. https://opendata.dwd.de/climate_environment/CDC/observations_germany/climate/10_minutes