The aforementioned issues lead to inherent uncertainty in the information required to improve grid reliability with further penetration of renewable generation. Since this uncertainty is poorly quantified, conservative grid management leads to curtailment of renewable power production (i.e., wind). Over the last year or two, there has been a rapid expansion in the deployment of solar power systems, especially at the smaller scale, which has added to the complexity of this situation.
To address these and related challenges, the Vermont Weather Analytics Center (VWAC) has been developed. It employs a system of coupled models to enable improved understanding of both production and demand. While each model attempts to deliver more accurate results, the models also quantify the inherent uncertainty in each respective component. The effort starts with weather, and builds upon the on-going work with IBM Deep Thunder, a state-of-the-art high spatial- and temporal-resolution forecasting system, customizable to meet the needs of specific weather-sensitive business decisions. It is based, in part, on the ARW core of the Weather Research and Forecasting (WRF) model. For Vermont, it is run operationally twice daily (initialized at 00 and 12 UTC) nested to 1-km horizontal resolution with high vertical resolution in the lower boundary layer for regional coverage for 72 hours. A number of model and remote sensing data sets are ingested to enable appropriate initial and boundary conditions. Three-dimensional variational data assimilation is performed around each analysis time using observations from the Vermont mesonet, which was developed at part of VWAC, as well as from MADIS and EarthNetworks WeatherBug mesonets.
Once each execution of the weather model is completed, the results are abstracted to include key variables at the appropriate temporal and spatial resolution. The variables include both direct model output as well as diagnostic fields derived from specialized post-processing. These data then permit execution in parallel of data-driven (i.e., via statistical and machine-learning) models to predict wind and solar power, and electricity demand. All of these models operate at a granularity that enables aggregation from the 1 km computational weather grid. Hence, at the finest scale, wind power is done at the turbine level, solar at each utility-scale facility and demand at the distribution substation. In addition, the demand model predicts solar power generation for distributed systems behind the meter (e.g., rooftop deployments), aggregated to the substation level. These models use training sets, which consist of both forecasts and hindcasts of the weather model, and historical power and other data from the utilities.
We will present an overview of the deployed capabilities along with the results to date and the overall effectiveness of our particular approach. We will also discuss on-going issues such as calibration of data and quantifying uncertainty as well as recommendations for future work.