STATISTICAL WIND POWER FORECASTING MODELS:RESULTS FOR U.S. WIND FARMS

摘要

Electricity markets in the United States are evolving. Accurate wind power forecasts arebeneficial for wind plant operators, utility operators, and utility customers. An accurate forecastmakes it possible for grid operators to schedule the economically efficient generation to meet thedemand of electrical customers. In the evolving markets some form of auction is held for variousforward markets, such as hour ahead or day ahead. In California, the Independent SystemOperator (ISO) filed a tariff for intermittent generators that uses independently developed windpower forecasts to help with resource selection in the hour-ahead market. The Federal EnergyRegulatory Commission (FERC) has recommended inclusion of this tariff in its Standard MarketDesign. This paper develops several statistical forecasting models that can be useful in houraheadmarkets that have a similar tariff. Although longer-term forecasting relies on numericalweather models, the statistical models used here focus on the short-term forecasts that can beuseful in the hour-ahead markets. The purpose of the paper is not to develop forecasting modelsthat can compete with commercially available models. Instead, we investigate the extent to whichtime-series analysis can improve on simplistic persistence forecasts. This project applied a classof models known as autoregressive moving average (ARMA) models to both wind speed andwind power output. The ARMA approach was selected because it is a powerful, well-knowntime-series technique and has been used by the California Independent System Operator in someof its forecasting work. Results are presented for operating wind farms in Iowa and Minnesota,and indicate that a significant improvement over persistence models is sometimes possible.