Analysis of winter weather conditions and their potential impact on wind farm operations

Severe weather conditions have two primary impacts on wind farm operations. The first relates to understanding potential damage to the turbines themselves and what actions are required to mitigate the effects. The second is recognizing what conditions may lead to a full or partial shutdown of the wind farm with sufficient lead time to determine the likely inability to meet energy generation committments. Ideally, wind forecasting suitable for wind farm operations should be of sufficient fidelity to resolve features within the boundary layer that lead to either damaging conditions or useful power generation. Given the complexity of the site-specific factors that effect the boundary layer at the scale of typical land-based wind farm locations such as topography, vegetation, land use, soil conditions, etc., which may vary with turbine design and layout within the farm, enabling reliable forecasts of too little or too much wind is challenging. A potential solution should involve continuous updates of alert triggering criteria through analysis of local wind patterns and probabilistic risk assessment for each location. To evaluate this idea, we utilize our operational mesoscale prediction system, dubbed “Deep Thunder”, developed at the IBM Thomas J. Watson Research Center. In particular, we analyze winter-time near-surface winds in upstate New York, where four similar winds farms are located. Each of these farms were built at roughly the same time and utilize similar turbines. Given the relative uncertainty associated with numerical weather prediction at this scale, and the difference in risk assessment due to the two primary impacts of severe weather, probabilistic forecasts are a prerequisite. Hence, we have employed ensembles of weather scenarios, which are based on the NCAR WRF-ARW modelling system. The set of ensemble members was composed with variations in the choices of physics and parameterization schemes, and source of background fields for initial conditions with horizontal grid resolutions in the one to two km range. In addition, the vertical grid structure was defined to ensure at least ten levels within the boundary layer and two from the bottom to the top of the turbine. This approach enables us to estimate the variability of winds at the farms and how it is distributed over the region. Further, we analyze the potential differences in structural risks at these farms during the 2009 winter season, and whether such differences in wind and weather patterns should be considered in choice of turbine design, installation and operations. We believe that this methodology can be extended to provide an estimate for mean annual energy production at a wind farm with the potential to improve the quality of siting and layout.