基于谱聚类和优化极端学习机的超短期风速预测.pdfVIP

39 卷 5 期 电 网 技 术 Vol. 39 No. 5 20 15 年5 月 Power System Technology May 2015 文章编号：1000-3673 （20 15）05-1307-08 中图分类号：TM 835 文献标志码：A 学科代码：470 4017 基于谱聚类和优化极端学习机的超短期风速预测 王辉，刘达，王继龙 （华北电力大学 经济与管理学院，北京市 昌平区 102206 ） Ultra-Short-Term Wind Speed Prediction Based on Spectral Clustering and Optimized Extreme Learning Machine WANG Hui, LIU Da, WANG Jilong (School of Economics and Management, North China Electric Power University, Changping District, Beijing 102206, China) ABSTRACT: The ultra-short-term wind speed prediction with 摘要：较高精度的超短期风速预测是并网运行风电场风电功 higher accuracy is the necessary premise and assurance for the 率预测预报系统建立和运行的必要前提及保证。由于风速影 establishment and operation of wind power forecasting system 响因素众多，具有较大的波动性和随机性，并具有高度的自 for grid-connected wind farm. Numerous factors impacting on 相关性，给传统的风速预测方法带来了极大的挑战。提出一 wind speed, which possess evident fluctuation property and 种基于谱聚类和极端学习机的超短期风速预测方法。该方法 randomness as well as high degree of self-correlation, bring 首先利用小波变换和主成分分析对风速数据进行去噪和降 huge challenge to traditional wind speed prediction method. A 维处理，剔除数据的不规则波动，有效降低数据维度；然后 spectral clustering (SC) and extreme learning machine (ELM) 分别应用谱聚类对小波变换后的各分解序列进行聚类分析， based ultra-short-term wind speed prediction method is 减少训练样本空间，提高样本有效性，降低计算复杂度；再 proposed. Firstly, the wavelet transform and principal 应用极端学习机对各分解序列分别进行训练，同时通过遗传 component analysis are utilized for the denoising and 算法对极端学习机输入权值、偏置等参数进行优化，确保各 dimensionality reduction of original wind speed data to get rid 分解序列输出最佳预测模型；最后将各分解序列预测结果相 of irregular fluctuation of original data to effectively reduce the 加得到最终预测结果。以某风电场实际数据进行的建模结果 dimensionality of original data; secondly,