ABSTRACT
Normal distributions do not adequately model the distribution of forecast errors in load and wind forecasts. Wind forecasts are created using statistical or physical models. Statistical models rely on historical wind data to predict future values based on trends and generally employ time series or neural network analysis methods. Physical models, often referred to as numerical weather prediction (NWP) models, use meteorological data to predict wind speeds that are converted to wind turbine output electrical power with a transfer function. Similar to load forecasting, the aggregation of wind plant forecasts over a large region tends to reduce forecast errors due to spatial smoothing of generation output. Wind forecast errors at different locations tend to cancel each other out to some extent. Note that the Midwest Independent System Operator (MISO) wind plants span much more area than those in Electric Reliability Council of Texas (ERCOT), and thus MISO benefits from greater geographical diversity in both wind and load forecasts.
