Using Reanalysis Data for the Prediction of Seasonal Wind Turbine Power Losses Due to Icing

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner
Monday, 3 February 2014: 11:15 AM
Room C114 (The Georgia World Congress Center )
Daniel G. Burtch, University of North Dakota, Grand Forks, ND; and G. L. Mullendore, D. J. Delene, and B. Storm

Handout (910.6 kB)

The Northern Plains region of the United States is home to a significant amount of potential wind energy. However, in winter months capturing this potential power is severely impacted by the meteorological conditions, in the form of icing. Predicting the expected loss in power production due to icing is a valuable parameter that can be used in wind turbine operations, determination of wind turbine site locations and long-term energy estimates which are used for financing purposes. Currently, losses due to icing must be estimated when developing predictions for turbine feasibility and financing studies, while icing maps, a tool commonly used in Europe, are lacking in the United States. This study uses the Modern-Era Retrospective Analysis for Research and Applications (MERRA) dataset in conjunction with turbine production data to investigate various methods of predicting seasonal losses (October-March) due to icing at two wind turbine sites located 121 km apart in North Dakota.

The prediction of icing losses is based on temperature and relative humidity thresholds and is accomplished using three methods. For each of the three methods, the required atmospheric variables are determined in one of two ways: using industry-specific software to correlate anemometer data in conjunction with the MERRA dataset and using only the MERRA dataset for all variables. For each season, a percentage of the total expected generated power lost due to icing is determined and compared to observed losses from the production data. An optimization is performed in order to determine the relative humidity threshold that minimizes the difference between the predicted and observed values. Eight seasons of data are used to determine an optimal relative humidity threshold, and a further three seasons of data are used to test this threshold.

Preliminary results have shown that the optimized relative humidity threshold for the northern turbine is higher than the southern turbine for all methods. For the three test seasons, the optimized thresholds tend to under-predict the icing losses. However, the threshold determined using boundary layer similarity theory most closely predicts the power losses due to icing versus the other methods. For the northern turbine, the average predicted power loss over the three seasons is 4.65 % while the observed power loss is 6.22 % (average difference of 1.57 %). For the southern turbine, the average predicted power loss and observed power loss over the same time period are 4.43 % and 6.16 %, respectively (average difference of 1.73 %). The three-year average, however, does not clearly capture the variability that exists season-to-season. On examination of each of the test seasons individually, the optimized relative humidity threshold methodology performs better than fixed power loss estimates commonly used in the wind energy industry.