改进的EMD及其在风电功率预测中的应用  

Improved Empirical Mode Decomposition and its Application to Wind Power Forecasting

作　　者：王鹏[1,2] 陈国初[1] 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：《控制工程》2011年第4期588-591,599,共5页Control Engineering of China

基　　金：国家自然科学基金项目(70773041);上海市教委重点学科(J51901);上海市教委科研创新重点项目(09ZZ211);上海市闵行区科技项目(2010MH169);上海市自然科学基金项目(11ZR1413900)

摘　　要：针对非平稳信号在经验模态分解(Empirical Mode Decomposition,EMD)过程中包络拟合时出现的过冲/欠冲问题,提出采用分段三次Hermite插值代替三次样条插值作为新的包络拟合算法。针对神经网络对非平稳性功率序列预测困难问题,采用EMD和神经网络相结合的方法对发电功率进行预测。使用改进的EMD对功率序列进行分解,降低序列的非平稳性,然后使用神经网络对各分量进行预测,最后通过叠加得到预测结果。利用该方法对东北某风电场进行功率预测,仿真结果表明,与其他方法相比,基于改进的EMD-ANN方法具有较高的预测精度。A envelopes fitting method is proposed based on the piecewise cubic Hermite interpolation,because the traditional cubic spline interpolation causes overshoots/undershoots for non-stationary signal in the process of empirical mode decomposition（EMD）.For the difficult forecasting problems of non-stationary power series using artificial neural network（ANN）,combined method of EMD and ANN is proposed.Power series can be decomposed into different series in order to reduce non-stationary sequence,and then ANN is used to forecast power by using each component.The prediction result is obtained through reconstructing.Using the data from a wind farm at the Northeast China,the simulation result shows that the prediction accuracy has significantly improvement compared with other methods.

关 键 词：经验模态分解 HERMITE插值 人工神经网络 风电功率 预测 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]