基于小波分解的风电场短期功率混合预测模型  

Hybrid Forecasting Model for Short-Term Power of Wind Farm Based on Wavelet Decomposition

作　　者：陈国初[1] 王鹏[1,2] 徐余法[1] 俞金寿[2] 

机构地区：[1]上海电机学院电气学院,上海200240 [2]华东理工大学信息科学与工程学院,上海200237

出　　处：《上海电机学院学报》2011年第3期163-168,共6页Journal of Shanghai Dianji University

基　　金：国家自然科学基金项目资助(60772006;70773041);上海市教育委员会重点学科资助(J51901);上海市教育委员会科研创新项目资助(09ZZ211);上海市闵行区科技项目资助(2010MH169)

摘　　要：为提高预测前1 h风电功率的精度,提出一种基于小波分解(WD)的人工神经网络(ANN)及最小二乘支持向量机(LS-SVM)相结合的预测方法。通过小波分解将功率和风速序列分解为不同频率的子序列,根据风机输出功率特点分析,对低频和高频子序列分别采用ANN法和LS-SVM法进行预测,最后通过重构得到预测结果。利用该方法对东北某风电场提前1 h的功率进行预测,实验结果表明:该模型具有较高的预测精度,与单纯的BP神经网络的模型相比,绝对平均误差从10.25%下降到5.62%。In order to improve the accuracy of predicting wind power one hour in advance,it is proposed to use a combination of artificial neural network（ANN） and least squares support vector machine（LS-SVM） based on wavelet decomposition.The power and speed series are decomposed into sub-sequences with different frequencies.Based on the characteristics of wind turbine output power,ANN and LS-SVM are built to make prediction using both low and high frequency sequences.The predicted results are then re-structured.Using the data taken from a wind farm in the northeast of China,the simulation results show that the wavelet-based hybrid forecasting model has high prediction accuracy,with the mean error reduced to 5.62% from 10.25% of simple BP neural network models.

关 键 词：小波分解 人工神经网络 最小二乘支持向量机 风电功率 预测 

分 类 号：TM614[电气工程—电力系统及自动化] TM71