Subhourly Wind Forecasting Techniques For Wind Turbine Operations PDF Download

Three models for making automated forecasts of subhourly wind and wind power fluctuations were examined to determine the models' appropriateness, accuracy, and reliability in wind forecasting for wind turbine operation. Such automated forecasts appear to have value not only in wind turbine control and operating strategies, but also in improving individual wind turbine control and operating strategies, but also in improving individual wind turbine operating strategies (such as determining when to attempt startup). A simple persistence model, an autoregressive model, and a generalized equivalent Markhov (GEM) model were developed and tested using spring season data from the WKY television tower located near Oklahoma City, Oklahoma. The three models represent a pure measurement approach, a pure statistical method and a statistical-dynamical model, respectively. Forecasting models of wind speed means and measures of deviations about the mean were developed and tested for all three forecasting techniques for the 45-meter level and for the 10-, 30- and 60-minute time intervals. The results of this exploratory study indicate that a persistence-based approach, using onsite measurements, will probably be superior in the 10-minute time frame. The GEM model appears to have the most potential in 30-minute and longer time frames, particularly when forecasting wind speed fluctuations. However, several improvements to the GEM model are suggested. In comparison to the other models, the autoregressive model performed poorly at all time frames; but, it is recommended that this model be upgraded to an autoregressive moving average (ARMA or ARIMA) model. The primary constraint in adapting the forecasting models to the production of wind turbine cluster power output forecasts is the lack of either actual data, or suitable models, for simulating wind turbine cluster performance.