At OCF we are building a state of the art national solar generation forecast that will be used by National Grid. By improving the solar generation forecast, we can help them to minimise the use of standby gas turbines, potentially leading to a substantial reduction in carbon emissions.
Build the best PV forecasting model you can. The model should forecast a PV system for the next 24 hours.
We would recommend that you use current data from other PV systems to help improve the forecast.
A PV dataset is available here or here. This dataset contains data from 1311 PV systems from 2018-01-01 to 2021-10-27. More data will be available soon.
We would recommend splitting the data into 2018 for training and 2019 for testing. Validation splits can be made how you wish in the training subset.
You can of course use other data, for example satellite data, or NWP data, but you are welcome to use other sources
For now lets use MAE, but perhaps we can explore a few other metrics that help differentiate a excellent model, from a good model
Please contact OCF if you like to contribute, or make a PR. We are interested in getting lots of people involved and building a great solution together.