Analysis of Short-term Wind Power Forecasting in the Northeastern United States

Power forecasting is vital for the scheduling and safe, stable operation of the electric power grid. Given the penetration of renewable generators, utilities are relying on power forecasting technologies to understand potential instabilities in the grid due to intermittency. Given the latter concern in Vermont, this paper details a short-term wind power forecasting solution that provides accurate forecasting results on prediction horizons up to 48 hours ahead by exploiting statistical modeling approaches based on the IBM SPSS modeler. The work leverages the IBM Hybrid Data-assimilation based Renewable Energy Forecasting (HyREF) technologies to support predictions for each turbine. To enable these forecasts, a database of historical data has been developed for training purposes from multiple sources such as power generation, SCADA, weather sensors on a meteorological tower and a weather model. The data are updated twice daily support operational forecasting. The weather model data are generated by IBM Deep Thunder, a state-of-the-art high spatial- and temporal-resolution forecasting system, and this model is customized to meet the needs of Vermont specific weather-sensitive business decisions. It is based, in part, on the ARW core of the Weather Research and Forecasting (WRF) model, nested to 1-km horizontal resolution with high vertical resolution in the lower boundary layer for regional coverage. To develop probabilistic wind power forecasts, statistical modeling methods such as CHAID tree, Artificial Neural Network, and Classification and Regression tree are combined by the SPSS modeler. Results of this capability deployed for a wind farm in Vermont will be presented and analyzed.