A study on the effects of auxillary data on the performance of an LSTM solar PV forecasting model
Short-term forecasting (nowcasting) of solar energy sources' outputs is an integral part of the successful decarbonization of energy grids. Without accurate prediction of solar power sources’ output, it is impossible to anticipate the amount of electricity available and provide a robust supply of energy to the consumers.
Currently, the majority of forecasting techniques rely on previous solar photovoltaic (PV) sources' outputs for prediction, or sometimes take into account auxiliary data such as weather or outputs of the nearby solar power sources. However, combinations of those are rare, and the contribution of auxillary data to the overall predictive power of the solution is not evaluated. This project aims to test the advantage of using any of those sources of data, as well as whether any significant improvement can be achieved when using them in combination.
- openclimatefix/uk_pv dataset was usead as the primary dataset of photovoltaic power generation
- ERA5 dataset was used as the source of the meteorological data, namely total cloud cover, skin temperature, and surface solar radiation
All analysis and data preparation performed can be found in data_prep_and_analysis
Done via optuna on UCL Myriad Cluster. Scrips used can be found in optuna_scripts
Done in Google Colaboratory on V100 GPU. Scripts used can be found in model_training
The authors acknowledge the use of the UCL Myriad High Performance Computing Facility (Myriad@UCL), and associated support services, in the completion of this work.