风电场风速预报集合订正方法的尝试性研究

以中国气象局风能太阳能资源评估中心风能预报系统(WINPOP)预报的甘肃HYZ和GHK两个风电场2012年大风月风速预报结果为基础,利用BP神经网络法(简称BP法)和自适应最小二乘法(简称LS法),采用两种训练样本构建方案对风电场风速预报进行订正试验,并尝试性使用神经网络集合方法对订正后的风速进行集合订正试验.结果发现:①经BP法和LS法等的直接统计订正及其集合订正后风电场风速预报效果明显优于WINPOP直接预报效果,且集合订正效果优于任何一种单一的统计订正;②与WINPOP直接预报相比,单一的统计订正及其集合订正都能较好地消除系统误差;③与单一的统计订正相比,集合订正具有更为明显的平均误差小、好日子率高、风速预报日变化幅度误差小等优势;④无论是HYZ风电场还是GHK风电场,随着预报时效的增加,各种预报(或订正)误差无显著差别.各种预报或订正方法的预报误差均随风速的增大而先减小后增大,随3h气压变量的增加而增大.

An Integrated and Revised Method of Forecasting Wind Speed for Wind Farms

Many statistical revised methods have been used in wind speed or wind power forecasting systems and research. However, these statistical methods have their own advantages. In this paper, an integrated and revised concept has been used which is more advanced and mostly used in weather forecasting. The revised test was done on wind speed for wind farms based on the observed wind speed and the forecast wind speed, temperature, air pressure, and wind direction from wind power forecasting system (WINPOP). Test methods included BP neural network, adaptive least square method and integrated methods. Test time: 1 March to 30 June 2012. Time resolution: every 15 minutes. Test place: HYZ wind farm and GHK wind farm in Gansu. Our results showed that wind speed forecasting by every single revised method is better than that of WINPOP and the error was reduced to 19.6%~30.8% and 23.4%~31.0% than that of WINPOP. The revised method error was 77%~93% and 78%~88% less than that of WINPOP for future 24 hours forecasting at HYZ and GHK. The error of integrated revised forecasting is less than that of WINPOP and every single revised forecasting. Error from the integrated revised method decreased to 40.5% and 36.8% and the percentage increased to 94% and 90% of revised error less than that of WINPOP on HYZ and GHK, respectively. Both the integrated method and single revised method remove system error of WINPOPPO remarkably. The integrated method can also remove daily variation error of WINPOP. There were no obvious changes for wind speed forecast error with prediction time increasing from 24h to 72h. The forecast error showed a decreasing trend for wind speed at(0~5)m/s and then increased for wind speeds greater than 5m/s.