Development of a wind turbine icing event database

Wind turbines operating in the presence of supercooled liquid water (icing) conditions suffer performance degradation and may even be shut down completely to prevent damage to the turbine itself. The loss of performance and potential down time translates directly to a loss of revenue for the operator, and potentially added costs from needing to purchase power on the open market to meet demand. Predicting the onset, likelihood, and severity of icing conditions at the locations of wind farms has the potential to offset some of these financial risks. However, several challenges exist in creating a robust system capable of producing accurate results at the height above the ground where wind turbines operate. Perhaps an even bigger challenge is generating a truth dataset at the height of the wind turbine, or more generally at the location of a wind farm.

Surface weather station (METAR) reports are notoriously error-prone at reporting freezing precipitation types at the surface and may not be representative of the conditions at the height of the wind turbine. Data from optical icing sensors installed at the height of the wind turbine can be inconsistent at identifying icing conditions, and are not available at all wind farms or on all wind turbines. An alternative method was developed to identify icing events by comparing the observed power output to what the expected power output should be based on the observed wind speed. This is weighted using the wind speed since low wind speed conditions are not as reliable at identifying power loss due to icing. This method must be combined with weather information since any manual reduction in turbine speed (curtailment) would look identical to the effect icing would have. Multiple sources of weather information will be compared with results of the power based icing metric including National Weather Service Zone Forecast Products, optical icing sensor data (when and where available), surface observations, and output from a point forecasting system known as Dynamic Integrated foreCast (DICast). These comparisons will be presented in an attempt to identify the strengths and weaknesses of using this method to develop a database of wind turbine icing events. This database will be used in the development of an intelligent forecasting system for the 19-44 hour time frame to assist wind turbine operators with decision-making.