Application And Researches Of The Mesoscale Model WRF In Wind Power Prediction

Due to random of wind speed and direction, the wind farm output power has the characteristics of volatility, intermittent and randomness. And forecast of wind power output is considered to be one of the most effective and economic ways to increase in power grid peak shaving capacity, improve the ability of acceptance of wind power, improve power system security and economy. Wind power prediction uses numerical weather prediction (NWP) and input variables of wind speed, wind direction. Then with prediction algorithm NWP variables are transformed into wind farm output power. Since there are many uncertainty in numerical weather perdiction, the accuracy of the NWP data significantly influences and restricts the wind power prediction accuracy.In this paper, we focus on to improve the accuracy and reliability of the NWP data, based on the WRF model, selected four areas in northern arid zone, a series of research is carried out:the optimum parameter configuration of the WRF model is got; based on this, got the distribution characteristics of model prediction error on the seasonal, monthly, daily and hourly time scales as well as under different weather background, the possible causes is also investigated; Finally, we try to apply the four dimension data assimilation, rapid update cycle and ensemble probability forecast technology to improve the accuracy of NWP data, and use the analog-based Kalman filter bias correction to correct the prediction error. We achieved good results in these works, and these techniques become the foundation of a real-time running business platform.Firstly, by using different boundary layer scheme of WRF model, we predicted the low level wind field in the four seasons over northwest China. The results show that:(1) the optimal solution of boundary layer in different seasons is different. Over Xinjiang, YSU scheme is the best in spring, MYNN2.5scheme is the best in summer; QNSE scheme is the best in fall and winter.(2) The performances of prediction vary in different seasons. Overall, the forecast of wind direction is very good; while for wind speed forecasting, it is better in spring and summer, the values is too large in fall and too small in winter.(3) For the four seasons the predicted wind speed is always smaller than the observed wind speed when it is large while larger when it is small. When the predicted values are too large in spring, fall and winter, the range of the observed values are<4m/s、<6m/s、<8m/s respectively; when the predicted values are too small in spring, summer, fall and winter the range of the observed values are>4m/s、>4m/s、6-10m/s、8-12m/s respectively.Secondly, by using the observed wind data over Xinjiang and Inner Mongolia, the performance of the WRF model in predicting the wind speed was tested in complex terrains. Results show that:(1) the prediction is better under the resolution of3km×3km than1km×1km.(2) slow wind region (0-3m/s), the prediction is the worst and it is just a little better in rapid wind region (≥25m/s). The prediction is good over the region where the wind speed is between3m/s and25m/s. Over Xinjiang the prediction is the best when wind speed is between12m/s and25m/s, while for Inner Mongolia it is best between3m/s and12m/s.Thirdly, by using WRF model we studied wind power forecast error and the condition of low level wind field and the possible causes over the arid region of northwest China. The results show that (1) the forecast error is always large when the observed wind speed is small while small when it is large.(2) On the daily scale, when the mean wind speed is less than2.5m/s or the anomaly is higher than15m/s, the model performance is worst; the prediction is the best between12:00-17:00and the worst between7:00-9:00; the prediction is better in cloudy day than in clear day.(3) The prediction error is closely related to the stability parameters. It is found that the simulation of wind shear is too small, which may be due to the simulation error in the high level wind speed. That the M-O length in the model is larger than the real one may be the source of the errors.In the end, by using the four dimension data assimilation, rapid update cycle and analog-based Kalman filter bias correction technology we decreased the prediction error of wind field. Based on this, the ensemble probability forecast is adopted and built a real-time business platform. Results as follows:(1) based on WRF model, the four dimension data assimilation, rapid update cycle and analog-based Kalman filter bias correction technology are applied to the numerical weather forecast platform and built the numerical weather forecast platform. The case of2012in Gansu Province shows that the application of the four dimension data assimilation can decrease the RMSE by17%and MAE by20%; rapid update cycle technology can decrease the MAE from3.63to3.03and increase the correlation coefficients from0.65to0.78; analog-based Kalman filter bias correction can decrease the RMSE from2.93to2.42and the MAE from2.28to1.82and increase the correlation coefficients from0.80to0.84.(2) Based on the improvements of the prediction error, considering the difference of the surface condition, an ensemble probability forecast system is established which contains26prediction member, by using the multi-boundary condition and multi-physical parameterization schemes. The preliminary probability forecast products is generated, which provide more accurate meteorological information for the electric power dispatching and decision-making.