基于多维时间序列和BP神经网络的短期风电功率预测  

Short Time Wind Power Prediction Based on Multidimensional Time-series and BP Neural Networks

作　　者：曹新莉[1] 王树朋[2] 

机构地区：[1]武汉工程大学电气信息学院,湖北武汉430205 [2]武汉大学电气工程学院,湖北武汉430072

出　　处：《陕西电力》2014年第4期19-23,共5页Shanxi Electric Power

基　　金：国家自然科学基金资助项目(51107098)

摘　　要：由于风电的高度波动性和随机性,大规模的风电功率预测已成为制约中国风电发展的瓶颈。提出一种针对小采样间隔的风电功率数据的多维时间序列BP神经网络预测模型。通过对原始风电功率序列进行处理得到不同时间维度的风电功率均值序列进而组成多维时间序列,采用改进的嵌入维最小预测误差法求取多维时间序列相空间重构时间延迟和嵌入维,利用重构相空间中预测点的近邻点建立BP神经网络预测模型。以实际风电场数据进行验证,证明了该模型可以有效处理风电功率预测问题,算法耗时减少了约9s,同时显著提高预测精度约18.94%。Due to the high volatility and randonmess of wind power,large-scale wind power prediction has become a bottleneck which restricts the development of China＇s wind power.This paper proposes a kind of multidimensional time series and BP neural network nrediction model aiming at small sampling intervals of wind power data.Different time dimension of the mean wind power sequence is ohtained by nrocessing the wind Power seqUence to create the multidimensional time series.The improved method of minimum embeddingd dimension is used to calculate the phase space reconstruction time delay and embedding dimension with multidimensional time series.BP neural network prediction model is established by using the neighbor prediction points in phase spaces.With the actual wind field data for validation,it is proved that the model can effectively predict wind power.The algorithm can reduce the time-consuming Of about 9 s and at the same time it can significantly improve the prediction accuracy of about 18.94%.

关 键 词：风力发电 风电功率预测 多维时间序列 BP神经网络 

分 类 号：TM614[电气工程—电力系统及自动化]