Forecast of the Wind Speed using the Regional Atmospheric Modeling System (RAMS) and Weather Research and Forecasting (WRF) models

Abstract

Physical, statistical models or a combination of both are used for the wind power prediction. Physical models considered meteorological and geophysical data to determine the value of the speed of the wind and with this power generation; statistical models, on the other hand, used historical data of electric generation. The latter integrated wind speed obtained from a numerical model. If the wind speed is not forecast within a range of acceptable error, power generation forecast will be affected in a critical way. This study presents the development of a hardware-software infrastructure to provide a short-term wind forecast, 4 times a day using the model Regional Atmospheric Modeling System (RAMS) and the Weather Research and Forecasting (WRF) 1 km resolution in an area of 1344 km2 located in the South of the Isthmus of Tehuantepec, Oaxaca. From the models are obtained datasets at the height of 80 m. Databases used as initial conditions and frontier models are data from the Global Forecast System (GFS) of the National Centers for Environmental Prediction (NCEP) for a period of one year. As the technique of adjustment models of prognosis numerical of weather (NWP) was implemented Kalman filter algorithm trying to eliminate systematic errors that are generated when modeling at levels close to the Earth's surface. As an option of statistical models were 3 models Autoregressive Integrated Moving Average (ARIMA) using historical data of wind speed. Forecast of the wind speed of all configurations was validated by comparing it with data measured at 80 m with a weather station located in the area, which belongs to the National Institute of electricity and clean energies (INEEL). Chai and Draxler [4] recommended using more than one metric to validate the models. The statistics were used in this study: mean absolute deviation (MAD), the mean absolute error (MAE) and the root of the mean square error (RMSE). The results show that the best model of forecasting for the period of 5 days is the WRF with an average RMSE of 2.48 and MAE average of 1.7. Forecasts 24 hours the best choice turned out to be the Kalman filter applied to the outputs of the RAMS model. This model shows the mean values of RMSE 1.74 and MAE of 1.32. Taking into account these results were operationally configured models and in a geographical information system provided the best forecasts 4 times a day every 6 hour. As future work in the short term is planned to make the forecast of wind and comparison of actual power generation with the forecasted for the wind turbine KWT300 (300 kW) located within the Regional Centre of Technology (CERTE) of the INEEL.