Photovoltaic output forecasting
Solar photovoltaic (PV) is rapidly becoming a significant source of humanity’s electricity. Fluctuations in solar PV output due to short-term events, like moving clouds, can have large impacts in areas with high solar PV penetration. This is particularly true where panels are geographically concentrated in industrial-scale PV farms. For this reason, significant effort has been made to forecast solar PV output using a variety of methods. Images of the sky contain a wealth of information, but this information is challenging to extract and use for reliable predictions. Recently, efforts have shifted to using machine vision systems to “read” the sky and make forecasts of PV panel output. Our group has developed a specialized convolutional neural network model named SUNSET (Stanford University Neural Network for Solar Electricity Trend) for 15 min ahead PV output forecast.
The following four projects have been published, with more research projects going on.
Nowcast: We explore the inference of solar panel output solely from concurrent local sky images with a convolutional neural network (CNN). This research is the first time that CNNs – and by extension deep learning – have been applied to predict solar panel output. We demonstrate that sky images are useful in inferring PV panel output, and CNN is a suitable structure in this application.
Forecast: We propose a specialized convolutional neural network (CNN) “SUNSET” to predict 15-min ahead minutely-averaged PV output. The model is characterized by its usage of hybrid input, temporal history and strong regularization. Optimal input and output configurations are explored and suggestions are given. The code base of this work is available on Github.
Data fusion: We systematically explore 28 methods of “fusing” the heterogeneous inputs, i.e., PV power output history and ground-based sky images, in our CNN, to ensure that due importance is given to each type of input. We also systematically explore the many hyperparameters related to model training and tuning. Limited resources preclude an exhaustive search.
Sky condition specific sub-models: We propose a two-stage classification-prediction framework for the nowcast task and compare it with the end-to-end SUNSET model we developed in previous research. The proposed framework first classifies input images into different sky conditions and then the classified images are sent to specific sub-models for PV output prediction.
Related Publications
- Nie, Y., Paletta, Q., Scott, A., Pomares, L., Arbod, G., Sgouridis, S., Lasenby, J., & Brandt, A. (2024). Sky image-based solar forecasting using deep learning with heterogeneous multi-location data: Dataset fusion versus transfer learning. Applied Energy, 369, 123467. https://doi.org/10.1016/j.apenergy.2024.123467
- Nie, Y., Zelikman, E., Scott, A., Paletta, Q., & Brandt, A. (2024). SkyGPT: Probabilistic ultra-short-term solar forecasting using synthetic sky images from physics-constrained VideoGPT. Advances in Applied Energy, 14, 100172. https://doi.org/10.1016/j.adapen.2024.100172
- Nie, Y., Li, X., Paletta, Q., Aragon, M., Scott, A., & Brandt, A. (2024). Open-source sky image datasets for solar forecasting with deep learning: A comprehensive survey. Renewable and Sustainable Energy Reviews, 189, 113977. https://doi.org/10.1016/j.rser.2023.113977
- Nie, Y., Zamzam, A., & Brandt, A. (2021). Resampling and data augmentation for short-term PV output prediction based on an imbalanced sky images dataset using convolutional neural networks. Solar Energy, 224, 341-354. https://doi.org/10.1016/j.solener.2021.05.095
- Nie, Y., Sun, Y., Chen, Y., Orsini, R., & Brandt, A. (2020). PV power output prediction from sky images using convolutional neural network: The comparison of sky-condition-specific sub-models and an end-to-end model. Journal of Renewable and Sustainable Energy, 12(4), 046101. https://doi.org/10.1063/5.0014016
- Venugopal, V., Sun, Y., & Brandt, A. (2019). Short-term solar PV forecasting using computer vision: The search for optimal CNN architectures for incorporating sky images and PV generation history. Journal of Renewable and Sustainable Energy, 11, 066102 . https://doi.org/10.1063/1.5122796
- Short-term solar power forecast with deep learning: Exploring optimal input and output configuration. (2019). Solar Energy, 188, 730-741. https://doi.org/10.1016/j.solener.2019.06.041
- Sun, Y., Szűcs, G., & Brandt, A. (2018). Solar PV output prediction from video streams using convolutional neural networks. Energy & Environmental Science, 11, 1811-1818. https://doi.org/10.1039/C7EE03420B