Probabilistic forecasting of aggregated generation for regional wind farms with geographical dynamic model

The interval estimation or a probabilistic density function of forecasted wind generation will provide valuable information for unit commitment problem with significant wind penetration. In this paper, we present a stochastic dynamical systems model for aggregated generation of regional wind farms combined with mesoscale numeric weather predicting and near surface field backtracking. An approach based on extended Kalman filter is used to forecast the probability of wind generation. A dynamic system based on MM5 and dynamic downscaling technology is established to forecast the wind speed of each wind farms in a region. The near-surface wind speed is forecasted through steps-ahead extended Kalman filter and the probability distribution of aggregated generation of wind farms is estimated and then the interval forecasts are obtained. The temporal and spatial correlations among regional wind farms are studied. Numerical testing results based on the actual data with 29 wind farms from the NREL and NCEP/NCAR Reanalysis datasets show that the improvement over the persistence model is more than 30% when the forecasting horizon increases to 6 hours.