Analysis of wrf model performance for short-term wind prediction and wind power applications

Recently, global climate change effects due to antropogenic pollutants, are much more obvious. This situation has caused increasing interest in clean energy resources like wind, solar, geothermal, tidal etc. Wind energy has become one of the most progressive clean energy around the world. Thus, especially in recent years, wind analyzes in a certain area before wind turbine or wind farm applications, widely take place in the literature. In addition, statistical analyzes and predictions are crucial for wind power studies. All these analyzes form the basis of determination of potential for regions where clean energy systems will be built. Due to high system installation costs, these preliminary studies are important in order to minimize the financial losses. In this study, data was taken from a measurement mast which was established in the area where the first national turbine will take place within a TUBITAK (The Scientific and Technological Research Council of Turkey) project called "Development of National Wind Energy Systems and Prototype Turbine Production". The measurement mast stands in the nortwest of İstanbul at 51 meters height from the sea level. Latitude and longitude of the mast is 41o 18` N and 28o 39` E. Temperature, pressure and wind data are measured at 81 meters, 80 meters, 65 meters, 40 meters and 20 meters on the mast. Measurement devices have been working properly through a solar panel that is mounted at 3 meters. The main purpose of the study is analyzing the wind characteristic of Terkos Region and providing base for the future studies by making short-term prediction about wind energy. In the first part of the study, exploratory data analysis was performed to summarize and visualize the characteristics of the data set covering the period between August 1, 2012 and August 1, 2013. Data from different levels of the measurement mast was used for obtaining probability distribution function. Then, Weibull distribution and Weibull parameters that are used to find the relative frequencies of wind speeds were determined. Then, prevailing wind direction and time series were plotted on the graphs by using Windographer. The prevailing wind directions of the region were identified as the northeast and southwest. Moreover, seasonal, monthly and daily time series were examined. Seasonal analyzes showed that wind speeds are higher in winter season than summer season. The highest wind speeds are respectively on October, November and February. Monthly analyzes indicated that wind speeds decrease towards the summer, and then rise again in the autumn. Besides, daily mean wind values changes between approximately 4 m/s and 7 m/s. In the next section, using the results obtained from the analysis, Wind Atlas Analysis and Application Program (WAsP) was performed in order to determine the wind potential of the region. Roughness map was derived by considering area topography and surface types and then given as a background map for WAsP. Spatial wind speed and wind power distributions were obtained and showed on the maps. These maps were compared with the same distribution maps obtained from a CFD model, MeteodynWT. The differences of the outputs from these two different model were examined. In the computations, WAsP has an assumption that the atmophere is stable. Contrary to this assumption, MeteodynWT can be operated under different atmospheric stability conditions. MeteodynWT decide the stability condition based on Monin-Obukhov length. Thus, atmospheric stability was investigated primarily before running the model by the vertical wind shear. The stability condition of the atmosphere for study period was determined as close to neutral, and the results were determined accordingly. The results on the maps were obtained for 65 m in both two models considering the turbines which have been planning to establish with 65 m hub height. Furthermore, because MeteodynWT is a CFD model, small changes can be included to the computations. Micro-scale events like turbulence can also be studied according to turbulent kinetic energy (TKE) budget equation. Hence, turbulence intensity map was also showed across the region. Digital map used in both WAsP and MeteodynWT covers 63 km2 area. The map was created by using grids that have 50 meters horizontal resolution. The contour lines on the map was shown with 3 m intervals. These contour lines reveal the topography of the study area. According to WAsP results, mean wind speeds on the area are between 5.13 m/s and 7.64 m/s on 65 m. On the same level, MeteodynWT results are similar to WAsP results. Mean wind speeds change from 5.20 m/s to 7.60 m/s over the region. Referring to the wind power density, it varies between 157 W/m2 and 525 W/m2 with reference to WAsP results. However, according to MeteodynWT outputs, wind power density values change between 189 W/m2 and 476 W/m2 throughout the region. An example of wind turbine application on the region, a 500 kW Enercon E40 type turbine was placed on the terrain and then wind energy production outputs were derived in both WAsP and MeteodynWT. In second designed, a wind farm that has 2.5 MWh of installed capacity of wind power, was applied on the terrain with five 500 kW Enercon E40 type turbines. Energy production outputs were calculated in two models again. Consequently, capacity factor was calculated as approximately 25% for both examples. Energy production of first application for the installed capacity of 500 kWh was calculated as 1104 MW per year in WAsP. This value was found as 1113 MW per year in MeteodynWT. For second wind farm design, these values are 5467 MW per year in WAsP and 5510 MW per year in MeteodynWT. The last part of the study consists of the short-term wind prediction by using modern mesoscale numerical prediction model, The Weather Research and Forecasting (WRF). The model has two dynamical core variants named nonhydrostatic mesoscale (NMM) and advance research (ARW). NMM is used for making operational forecasts while ARW is used for both meteorological research and numerical weather prediction. In this study, WRF-ARW version 3.3.1 was used. 3 day periods were chosen in order to obtain wind predictions on February and March. The predictions were achieved in case of one hour outputs. The model was run with 6 different initial conditions including microphysics, cumulus parametrization, planetary boundary layer (PBL) and atmospheric radiation physics schemes. The model results were downscaled to the latitude and longitude where measurement mast stands in Terkos, and then compared to the observation data. Firstly, for February 1-4 and March 1-4 results were obtain with 6 different parametrization of WRF. NCEP Final Operational Model Global Tropospheric Analyses data set with 6-hour interval, was used as initial data. Domain configuration was set up with 3 nested domains with 30 km, 10km and 3 km horizontal resolution, respectively. The model includes 28 vertical levels. Coarser domain covers an area between 33° N and 49° N latitudes and 19° E and 39° E longitudes where contains Turkey and some parts of Eastern Europe. The middle domain involves Marmara region located in the northwest of Turkey. The inner domain has an area including Terkos region. Errors for each parametrizations on each domains were determined by RMSE. On the purpose of reducing the errors, artificial neural network (ANN) was performed by using WRF-ARW results with different initial conditions. WRF results showed that, focusing on the study area with higher resolution provides that predictions converge to the observations. Another result is that where wind speeds are high, predictions are more consistent with the observations. Given the initial conditions, parametrizations indicated as "WRF-1" in the study, has closer prediction results to observation data. In addition, in the inner domain, RMSE values are less than other domains due to higher spatial resolution. However, predictions have high errors that cannot be underestimated for all situations. Thus, in order to have more consistent results, using micro-scale prediction models may be more applicable especially in wind predictions. Because, wind is quite variable even very short time. This situation makes wind predictions hard when comparing with more stabile variables, such as temperature. ANN results have made the corelation between model results and observations higher, so that reduced the errors of the model results. ANN study has showed that using WRF-ARW model results without post processing may lead to errors or inconsistent results. Thus, evaluating the model results in order to make more correct them in some methods is crucial, especially studies made for economical purposes. As a result, physical schemes based on different calculations and assumptions cause different model results. On the other hand, combination of consistent aspects of model results obtained from different physical schemes by some methods like ANN, can improve the correlations between forecasts and observations and reduce the errors. As a summary, in a wind energy or wind analysis study, the general characteristics of local winds and its effects should be examined in detail. For energy production in a selected area, appropriate turbine selection and the deployment of the turbine in the right place are important issues both in terms of using the natural sources efficiently and in terms of provision of financial gains. For the same purposes, short-term wind forecasts are also important. Another and the most important issue is having proper measurement systems both for analyzing the winds locally, and evaluating the success of model results by comparing them with reliable observation data. Thus, in this study, general and detailed information about local winds of Terkos Region were obtained by using measurements on the area, and then forecast performation of WRF model was evaluated with including the correction of different model forecasts by ANN method with a view to shed light on the establishment of wind energy systems which may locate on Terkos Region.