Normalising the target variable of power by irradiance to try and forecast efficiency instead
Detailed Description
Experiment to run on PVNet, assuming Power = Efficiency * Irradiance, we can try and model/forecast efficiency by normalising by one of the irradiance values we have in UKV or ECMWF NWPs and then running PVNet training
Context
This would be an interesting experiment to run to see what the model is learning and how this output differs when unnormalised compared to what we forecast for power. Inspired by some work @mattw294 has done on modelling efficiency to help PV power forecasts
Possible Implementation
Normalise the y, target variable, of power by the UKV dswrf downwards short-wave radiation flux value at the centre of each spatial window and then run PVNet training
Normalising the target variable of power by irradiance to try and forecast efficiency instead
Detailed Description
Experiment to run on PVNet, assuming Power = Efficiency * Irradiance, we can try and model/forecast efficiency by normalising by one of the irradiance values we have in UKV or ECMWF NWPs and then running PVNet training
Context
This would be an interesting experiment to run to see what the model is learning and how this output differs when unnormalised compared to what we forecast for power. Inspired by some work @mattw294 has done on modelling efficiency to help PV power forecasts
Possible Implementation
Normalise the y, target variable, of power by the UKV
dswrf
downwards short-wave radiation flux value at the centre of each spatial window and then run PVNet training