Improving NWP Forecasts for the Wind Energy Sector

Abstract

Weather forecasting has traditionally been primarily used in the energy industry to estimate the impact of weather, particularly temperature, on future electrical demand. As a growing proportion of electricity generation comes from intermittent renewable sources such as wind, weather forecasting techniques need to be extended to this highly variable and site-specific resource. We demonstrate that wind speed forecasts from Numerical Weather Prediction (NWP) models can be significantly improved by implementing a bias correction methodology. For the study presented here, we used the Australian Bureau of Meteorology (BoM) MesoLAPS 5 km limited domain NWP model, focused over the Victoria/Tasmania region of Australia. The site for this study is the Woolnorth wind farm, situated in north-west Tasmania. We present a comparison of the accuracy of uncorrected hourly NWP forecasts and bias-corrected forecasts over the period March 2005 to May 2006. This comparison includes both the wind speed regimes of importance for typical daily wind farm operation, as well as infrequent but highly important weather risk scenarios that require turbine shutdown. In addition to the improved accuracy that can be obtained with a basic bias correction method, we show that further improvement can be gained from an additional correction that makes use of real-time wind turbine data and a smoothing function to correct for timing-related issues that result from use of the basic correction alone. With full correction applied, we obtain a reduction in the magnitude of the wind speed error by as much as 50 % for ‘hour ahead’ forecasts specific to the wind farm site.