基于样本卷积交互网络的风电场集群短期功率预测  

SHORT-TERM POWER PREDICTION FOR WIND FARM CLUSTER BASED ON SAMPLE CONVOLUTION INTERACTION NETWORK

作　　者：朱国鹏 向玲[1] 范文振 吴俊 李跃文 胡爱军[1] Zhu Guopeng;Xiang Ling;Fan Wenzhen;Wu Jun;Li Yuewen;Hu Aijun(Department of Mechanical Engineering,North China Electric Power University,Baoding 071003,China;Luneng New Energy(Group)Co.,Ltd.,Inner Mongolia Branch,Hohhot 010010,China)

机构地区：[1]华北电力大学(保定)机械工程系,保定071003 [2]鲁能新能源(集团)有限公司内蒙古分公司,呼和浩特010010

出　　处：《太阳能学报》2025年第1期158-167,共10页Acta Energiae Solaris Sinica

基　　金：国家自然科学基金(52075170,52175092)。

摘　　要：为保障风电集群安全运行和优化区域电网调度,提出一种基于样本卷积交互网络(SCINet)的风电场集群短期功率预测方法。首先引入能量熵(EE)、变分模态分解(VMD)方法对功率序列进行处理,然后对平稳序列和非平稳序列分别使用SCINet、自回归滑动平均模型(ARMA)进行预测,最后将模型输出结果重构获得最终功率预测结果。算例1以中国东北某150MW大型风电场实测数据为例进行模型构建和预测分析,结果表明模型在功率序列特征挖掘方面具有明显优势,且预测精度较高。算例2以西北某298.5 MW风电场集群功率数据对所提方法进行验证,验证结果显示,该方法泛化性好,与目前风电场集群功率预测常用方法相比性能更好、计算效率更高,可为风电场集群功率预测提供参考。To ensure the secure and efficient operation of wind farm clusters and optimize regional grid dispatch,a novel methodology for predicting the power output of wind farm clusters based on the Sample Convolution Interaction Network(SCINet)is proposed.The fundamental principles of Energy Entropy(EE)and Variational Mode Decomposition(VMD)to process power sequences with high precision is integrated.Within this predictive framework,the Sample Convolution Interaction Network(SCINet)is employed to forecast stationary sequences,while the Auto Regressive Moving Average(ARMA)model is applied to non-stationary sequences.Subsequently,the outputs of the model are meticulously reconstructed to produce the final prediction results.In the first case study,empirical data from a 150 MW large wind farm located in northeast China are utilized for model development and comprehensive prediction analysis.The results demonstrate the significant advantage of the proposed model in effectively extracting distinctive features from power sequences,thus substantiating its superior predictive accuracy.In the second case study,the methodology is validated using power data from a 298.5 MW wind farm cluster in northwestern China.The validation results affirm the model′s robust generalization capability.Compared to conventional methods commonly used for wind farm cluster power prediction,the proposed approach not only outperforms existing methods but also exhibits enhanced computational efficiency.Consequently,it provides a valuable reference for accurate power prediction in similar wind farm clusters.

关 键 词：风功率 预测 风电场 信号处理 变分模态分解 卷积 

分 类 号：TM614[电气工程—电力系统及自动化]