University of Twente Student Theses
Encoder-Forecaster-Decoder: a Modular Deep Learning Framework for Cloudage Forecasting
Bijl, M.H. de (2020) Encoder-Forecaster-Decoder: a Modular Deep Learning Framework for Cloudage Forecasting.
|
PDF
8MB |
| Abstract: | Solar Team Twente is always looking to increase their performance in the world solar challenge. An essential part of their race strategy is the weather model, which is tasked to forecast the surface solar irradiance (SSI). The current weather model lacks detail in both spatial and temporal resolution. SSI forecasts based on geostationary satellite observations of cloudage do have the desired resolution. Since cloudage is highly correlated with SSI, forecasting cloudage is an accurate method for forecasting SSI. Based on state-of-the-art deep learning models, the modular encoder-forecaster-decoder framework is proposed to forecast cloudage. The models that make up the elements of this framework are determined by experimentation to be the SegNet model for the encoder and decoder, and the TrajGRU cell as the forecaster. The forecasting performance of the novel model, together with the TV-L1 optical flow method, the ConvLSTM model, and the TrajGRU model is measured for different forecast horizons on a data set containing cloud masks. In the end, it was determined that the TrajGRU model performs the best. The proposed framework lacks the ability to forecast due to inherent flaws in its design. |
| Item Type: | Essay (Master) |
| Clients: | Alten Nederland, Apeldoorn, Netherlands |
| Faculty: | EEMCS: Electrical Engineering, Mathematics and Computer Science |
| Subject: | 54 computer science |
| Programme: | Computer Science MSc (60300) |
| Link to this item: | https://purl.utwente.nl/essays/82522 |
| Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Show download statistics for this publication
Show download statistics for this publicationRepository Staff Only: item control page