基于小波阀值降噪和BP神经网络的超短期风电功率预测  

Ultra Short-term Wind Power Forecasting Based on Wavelet Threshold Denoising and BP Neural Network

作　　者：刘新东[1] 陈焕远[1] 佘彩绮[1] 

机构地区：[1]暨南大学电气信息学院,珠海519070

出　　处：《世界科技研究与发展》2011年第6期1006-1010,共5页World Sci-Tech R&D

基　　金：国家自然科学基金(51007030);中央高校基本科研业务费专项资金(21611420);大学生创新实验性项目(101055937)资助项目

摘　　要：针对超短期风电功率预测问题,考虑了风电场复杂的噪声背景和风电功率的波动性,提出了一种基于小波阀值降噪-BP神经网络的超短期风电功率预测方法该方法采用近似对称光滑的紧支撑双正交小波db4(Daubechies函数)作为小波基,通过多分辨分析的Mallat算法对历史时序风电功率数据进行3尺度分解。根据Donoho阀值法对各层小波系数进行软阀值降噪处理,再通过小波逆变换重构历史时序风电功率,由BP神经网络对其进行训练,预测目的风电功率序列。仿真算例将该方法与普通BP神经网络方法进行了对比,比较结果证明其预测精度优于后者,具有很好鲁棒性和降噪性能,适用噪声复杂的风电场超短期风电功率在线预测。For ultra short-term wind power forecasting problem, considering the complex noise background of the wind farm and the volatility of wind power, a method based on Wavelet threshold denoising and BP neural network is proposed. The approximate symmetry smooth compactly supported biorthogonal wayelet db4 （Daubechies function） is adopted as the wavelet bases. The muhiresolution analysis Mallat algorithm is used to decompose the historical time series wind power data to 3 scales. According to the Donoho threshold, the noise of each wavelet coefficient is reduced by soft-threshold. Then the historical time series wind power is reconstructed by Wavelet inverse transform. And the BP neural network is trained to forecast the target wind power sequence. This method and the common BP neural network are contrasted in the simulation. The comparative results prove its forecasting accuracy is higher than the latter. It has good robustness and noise reduction performance ,suitable for the ultra shortterm wind power online forecasting in the wind farm with complex noise background.

关 键 词：风电功率 超短期预测 小波分析 Donoho阀值 BP神经网络 

分 类 号：TM614[电气工程—电力系统及自动化] TP183[自动化与计算机技术—控制理论与控制工程]