Rye microgrid historical weather forecasts and stochastic scenarios

This datasets connects historical weather forecasts from the Norwegian Meteorological Institute (met.no) and historical observations from Rye microgrid (https://doi.org/10.5281/zenodo.4448894).

Each csv file represents a historical weather forecast for approximately 60 hours ahead. Each csv-file also contains the corresponding observations in the same time interval. Finally, the files also contain load, wind generation and solar PV generation predicitons.

The predictions are generated using gradient boosting. The predictions ending with "_ls" are based on least square. The predictions ending with "_quantile_i" represent a quantile prediction. For example, "wind_quantile_2" means that there is a 20% probability the wind will be less than this value.

The gradient boosting predictor has been trained to predict the wind power, solar power and load using the explanatory variables below:

Solar PV: Cloud area fraction, initial production, clear sky production and forecast look-ahead time

Wind power: wind speed, wind direction, wind power converted from wind speed forecast, initial production and forecast look-ahead time

Load: hour of day, month of year