You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Currently the window size to generate the recent_power feature is hardcoded in the recent_history.py file. Currently this value is to 30 minutes: recent_power_minutes = 30here:
Hence, if PV data has been used that has a 5 minutely frequency, the last 6 values would be used (assuming no PV lag) to generate this recent_power feature. This is generally okay but for cases where we want to maximise the models performance for shorter horizon duration forecasts, we loose information by averaging out the recent power over a larger time than the forecast horizon duration. https://github.com/openclimatefix/pv-site-prediction/blob/main/psp/models/recent_history.py#L519
A quick solution is to hardcode the value in recent history, but this will affect how other models run.
A more robust solution is to make this parameter configurable in the experiment config, but by default is set to 30.
Another solution could be to pass in more recent_power features into the model.
The text was updated successfully, but these errors were encountered:
Currently the window size to generate the recent_power feature is hardcoded in the recent_history.py file. Currently this value is to 30 minutes:
recent_power_minutes = 30
here:Hence, if PV data has been used that has a 5 minutely frequency, the last 6 values would be used (assuming no PV lag) to generate this recent_power feature. This is generally okay but for cases where we want to maximise the models performance for shorter horizon duration forecasts, we loose information by averaging out the recent power over a larger time than the forecast horizon duration.
https://github.com/openclimatefix/pv-site-prediction/blob/main/psp/models/recent_history.py#L519
The text was updated successfully, but these errors were encountered: