기상수치모델을 이용한 풍황 및 출력변동성을 고려한 풍력발전량 예보

Abstract

A wind turbine must be regularly maintained in order to manage a wind turbine by a automatic control system. When workers perform a maintenance of a wind turbine, it is necessary to secure their safety. In addition, they should try to minimize the amount of loss while not driving a wind turbine. For those condition, weather information are provided such as the wind speed including a power production in wind turbine. The domestic electric power market is bidding competition based on the supply capacity of power generation companies. There is a difficulty in providing available supply capacity for bidding competition, and the domestic renewable energy power capacity is not proceeding to bid competition. However, since the manager of the power market affects the power supply capacity of the other generation source as much as the power supply of the renewable energy supplied, it can not be ignored when the price development plan is established. Uncontrolled power production in wind farms should be informed of uncertainty for risk analysis of estimated power generation. This study proposes a real time wind power generation forecasting system using WRF model and provide variation information about power production forecast using Monte Carlo Simulation. The conceptual design and detailed design were established for real-time wind power forecasting system. The conceptual design determines the purpose of the forecast, the forecasting time, the forecasting resolution, the forecasting target and the forecasting method. The detailed design consisted of hardware and software installation, hardware performance test, WRF model optimization, statistical model, and forecast schedule. The forecasting purpose was the operation plan of the wind turbine, and the forecasting time was selected as the 3-day forecast and the 14-day forecast. The forecast resolution is 1 hour and 1 day, and the target is a single wind turbine, Hangwon no. 3. WRF model, numerical weather prediction model, was selected. The hybrid parallel computing(MPI + OpenMP) was selected in order to install the WRF model and maximize the performance of the model. The UV method, which is a process of decomposing wind speed and wind direction into u-speed and v-speed, is proposed. The prediction accuracy of wind speed and wind direction was examined by implementing the UV method to a regression model and a neural network model. When the data of the wind speed and direction are used without the UV method, the wind direction was adjusted in the wrong direction only considering the deviation. However, when the u-speed and v-speed are used with the UV method, the wind direction is adjusted in consideration of the directionality. The variation of the wind speed in the WRF model was calculated based on the RMSE results of the AWS data. The variation of a power curve reflected the proposed variation estimation formula and the variation of the air density. This formula is verified by comparing with the measured data. The accuracy of power production for the medium-term forecast was 180.32 ㎾h for MAE and 245.61 ㎾h for RMSE, and IOA for data consistency was 0.71. The accuracy of the long-term forecast was 4.28 ㎿h for MAE, 5.64 ㎿h for RMSE, and 0.51 for IOA. The variation forecast of the medium-term showed a forecast accuracy of 76.7%. The hourly forecast variation ranged from 58.8% to 94.7% during the three-day forecast period. The variation forecast of the long-term showed a forecast accuracy of 80.8%. The daily forecast variation ranged from 62.1% to 92.9% during the 14-day forecast period. The forecast variation of long-term forecasts was more than 85% accurate by the 6th day, and accuracy declined sharply from 7th.