Generating an ensemble of hourly renewable energy from a suite of daily weather parameters sets for energy system modelling: a two-step modelling approach

Sophisticated modelling of renewable energy (RE) production at a hourly time step is starting to produce valuable multi-decadal time series for the evaluation of capacity factors for specific climate time series and RE deployment scenarios. This enables an adequate representation of the variability in future energy generation in scenarios with large shares of RE into the electricity grid. Our work proposes a data-driven approach to extrapolate the relationships between weather and hourly RE capacity factors (CFs) observed in existing datasets to other climate conditions and RE deployment scenarios. In particular, many climate ensembles used by other sectors (food, water, etc.) have a daily time resolution, so how do we project plausible hourly CF time series to these ensembles for multi-sector studies? We describe a data-driven model generating hourly RE CFs for climate ensembles containing daily weather parameters time series (called the candidate ensembles thereafter), based on the observed correlation between hourly wind speed, solar irradiance (and other relevant variables) and corresponding wind and solar CFs in an existing dataset (called the validated dataset thereafter). We use a two-step approach. First, we use daily aggregated values of weather variables and RE CFs from the validated dataset to generate daily RE CF time series in the candidate ensembles. Second, we disaggregate the daily RE CFs data to hourly time resolution using the hourly patterns observed in the validated dataset.  To validate our approach, we use several different splits of the historical time series in the validated dataset into two separate datasets used for training and validating the model. For each validation dataset, we construct hourly CF time series from daily aggregated weather data, and we verify that the hourly RE CFs generated by the model statistically agree with the validated hourly RE CFs time series. We illustrate our approach with a validated dataset of Great Britain's (GB) historical weather and CF time series, and a 100-member ensemble of future daily weather parameters (weather@home2 (W@H2)) as the candidate dataset. This leads to the production of an ensemble of hourly-resolved RE CFs for two periods of 2023 to 2049 (near future) and 2070 to 2099 (far future). The approach can be used to represent the variability in energy generation for any climate ensembles, and could be applied for energy system modelling in industry and academic research.