An advanced model for the short-term forecast of wind energy

Abstract

A novel short-term wind energy forecasting method, which is being developed under the University of Brunei Darussalam - International Business Machines (UBD-IBM) renewable energy modeling initiative, is described in this paper. The paper starts with a brief review on the existing forecasting methods. Prediction models based on the physical (derived from Numerical Weather Prediction models) and Time Series approaches are discussed. The prediction errors under these methods are described and the need for a reliable forecasting system is emphasized. This is followed by a detailed discussion on the UBD-IBM approach. The baseline of the proposed forecasting system is the IBM Deep Thunder. Deep thunder is a highly modified version of the Regional Atmospheric Modeling System (RAMS). This high resolution Weather forecasting system can predict the local Weather variations on a 'day ahead' basis, at high accuracy levels. For a given wind farm site, specific Deep Thunder models could be developed and calibrated using the surface measured wind data. This enables the Deep Thunder to predict the wind profile at the wind farm locations over shorter time scales. Once the wind spectra at the reference height are obtained from the Deep Thunder, terrain characteristics of the wind farm site are introduced to the model using the geotropic drag law. This will further be extended to the specific turbine location by considering local variations in surface roughness, orographic effects etc. The wind farm wake effect is introduced to the model at this stage. Velocity deficit experienced by the downstream turbines due to the presence of the upstream ones would be modeled. A semi empirical approach, based on the conservation of momentum, would be adopted. With these procedures, the wind velocity, 'actually felt' by individual turbines in the farm could be predicted. From this predicted wind velocity at the hub height, the expected energy yield from individual turbines over a given time period is estimated. For this, the probability density of these wind speeds within the 'look-ahead period' and the power curve of the turbines are combined and integrated over the functional velocity range of the wind generator. The total energy expected from the wind farm can be computed by adding the energy yields from individual turbines.