Analysis of the Renewable Energy Assessment Programs RETScreen and System Advisor Model (SAM) - Wind Energy Model Predictions Comparison with Measured Operational Data

Abstract

In this study, the wind energy modules of the renewable energy assessment programs RETScreen and System Advisor Model (SAM) were examined, and their predictions compared to measured operational data. Both of these programs have been used in teaching energy infrastructure at the University of Washington. It is of interest to see how well they perform, since validation and similar research have been limited to date. The programs have integrated and associated web-based weather databases and, therefore, a preliminary assessment can be performed in the absence of onsite wind speed measurements. Operational data from a wind farm in the United States, which included electricity production and availability for a five-year period, as well as wind speed measurements from an onsite meteorological (MET) tower, were compared to the wind speed and AEP (Annual Energy Production) predictions of the models. Model predictions were made with and without accounting for energy losses. One standard deviation in predicted AEP and wind speed based on variation in the input parameter values was used as a measure of uncertainty. The wind farm is located in a complex topographical area, which increased the complexity of the comparison. The main conclusions, listed for each program, are the following: RETScreen: - The weather database associated with RETScreen has too coarse a spatial resolution to be accurate for a given site located in a complex topographical area. Low wind speeds obtained from the database (i.e. 25% - 32% lower than MET data from an onsite weather station), lead to an underestimation of the AEP in the range of [-34%; -45%], assuming no losses and [-47%; -56%], when accounting for losses. - Using a wind resource map from NREL (National Renewable Energy Laboratory) gives a better estimate of the wind speed at the site, in the range of [-3.4%; +4.5%] difference compared to the MET data. Consequently, the AEP was predicted more accurately in the range of [+25%; -8%], assuming no losses and [+1%; -27%], when accounting for losses. - For sites in complex terrain, it is recommended that NREL wind resource maps are used rather than the associated weather databases. However, this estimate should be considered to be very rough as the maps are given in increments of 0.5 m/s and the AEP is, therefore, predicted over a large range. - It is recommended that the shape factor and shear coefficient are defined as ranges rather than single values based on how much impact the determination of those has on the predicted AEP. SAM: - Two weather files at locations closest to the existing wind farm were selected as representative of the wind farm and the AEP predictions were performed for both. The annual average wind speed was 6.73 ± 0.06 m/s and 7.21 ± 0.22 m/s respectively, compared to 6.93 m/s wind speed at the MET tower. - For the closest site, the predicted average AEP is 13.3% higher compared to the average wind farm AEP and the range is [32%; -2%] ignoring losses. When accounting for losses, the difference range is [7%; -22%] and -8.7% when averages are compared (i.e. the average AEP in SAM has an 8.7% underestimation compared to the wind farm data). - Based on this case study, the AEP predictions in SAM are quite good (in the range of roughly [+10% to -30%] compared to the wind farm data) and using only one weather file to represent a whole wind farm is appropriate for rough AEP estimation. However, based on only one case in this study, it is not possible to generalize about the accuracy of the program and further studies would need to be conducted. One important limitation of this analysis is that the operational wind farm data and the SAM predictions do not cover the same period, resulting in uncertainty of the AEP comparison. However, it should also be noted that the predicted AEP from SAM does not vary significantly over the years when data was available. It is recommended that additional research will be conducted to eliminate this uncertainty. It would also be of interest to compare the models to operational data from a site in a flat terrain.