基于ARMA的风电功率预测
2016-07-09惠小健王震张善文贺海龙
惠小健 王震 张善文 贺海龙



摘 要: 风电场风电功率预测对优化电网调度,提高风电场容量系数具有重要意义。对采样时间为15 min的风电功率时间序列建立自回归移动平均模型,并对风电场输出功率分别进行短期和中长期预测,同时分别分析了4台风电机组和58台风电机组的汇聚对预测结果的误差影响等。研究结果表明,利用ARMA模型在预测短期及中长期风电功率时的日前预测平均相对误差为0.087 1,实时预测误差为0.15,同时4台风电机组和58台风电机组的汇聚的平均相对误差为0.293 1和0.194 3,风电机组在集中开发方式下风电功率预测误差减小。
关键词: 风力发电; ARMA; 风电功率预测; 风电机组
中图分类号: TN925?34; TM71 文献标识码: A 文章编号: 1004?373X(2016)07?0145?04
Abstract: Wind power forecast of wind power plant is very important to optimize the power grid dispatching and improve the coefficient of wind power plant. The auto regressive moving average (ARMA) model of wind power time series was established, whose sampling time is 15 min. The short?term and mid?long?term forecast for the output power of the wind power plant are conducted. The error effect of 4 wind turbines and 58 wind turbines on the forecast results is analyzed respectively. The research results show that the current average relative error is 0.087 1 when the established ARMA model is used to forecast the short?term and mid?long?term wind power, the real?time forecast error is 0.15. The average relative errors of 4 wind turbines and 58 wind turbines are 0.293 1 and 0.194 3 respectively. The prediction error of wind power is reduced while the wind turbines in concentrated development way.
Keywords: wind power generation; ARMA; wind power forecast; wind turbine generator
0 引 言
对风电场的发电功率进行尽可能准确的预测,是风力发电并保证电力系统安全可靠运行的一项长期研究课题。当前,从时间角度来讲,风电场输出功率的预测方法有超短期预测(数分钟);短期预测(数小时或数天),中长期预测(数周或数月)[1]。对于风电功率预测,文献[2]利用非参数回归模型得到风电功率的点预测值。文献[3]对短期风电功率预测提出了一种包含纵向误差,横向误差,相关因子与极端误差等在内的综合评价方法。文献[4]采用基于多层前馈人工神经网络(BP?ANN)的间接预测法对超短期的风电场输出功率进行预测。文献[5?9]也分别提出了主成分?遗传神经网络,小波?BP神经网络,人工神经网络方法,时间序列分析方法,卡尔曼滤波法……
