融合迁移学习与CGAN的风电集群功率超短期预测  

Fusion of Transfer Learning and CGAN for Ultra Short-term Power Prediction of Wind Power Clusters

作　　者：周军[1] 王渴心 王岩 ZHOU Jun;WANG Kexin;WANG Yan(School of Electrical Engineering,Northeast Electric Power University,Jilin 132012,China;Jilin Power Supply Company,State Grid Jilin Electric Power Company,Jilin 132012,China)

机构地区：[1]东北电力大学电气工程学院,吉林132012 [2]国网吉林省电力有限公司吉林供电公司,吉林132012

出　　处：《电力系统及其自动化学报》2024年第5期9-18,共10页Proceedings of the CSU-EPSA

基　　金：吉林省科技发展计划资助项目(20230203033SF)。

摘　　要：针对可再生能源不确定性导致电力系统消纳能力不足的问题,提出一种基于条件生成对抗网络与迁移学习融合的风电集群功率超短期预测方法。首先,分析了风电集群功率预测样本模式的不均衡性以及导致的神经网络预测误差偏移现象;其次,构建了条件生成对抗网络修复不均衡问题;最后,采用迁移学习结合时间卷积网络构建了风电集群功率超短期预测模型。测试结果表明,所提方法能够显著提高风电集群功率超短期预测精度。In response to the issue of insufficient absorption capacity of power system caused by the uncertainties in renewable energy,an ultra short-term power prediction method for wind power clusters based on the fusion of conditional generative adversarial network(CGAN)and transfer learning(TL)is proposed in this paper.First,the imbalance of sample modes in wind power cluster power prediction and the resulting deviation of neural network prediction errors are analyzed.Second,a CGAN is constructed to repair the imbalance problem.Finally,the combination of TL and temporal convolutional network is used to construct an ultra short-term power prediction model for wind power clusters.The test results show that the proposed method can obviously improve the ultra short-term power prediction accuracy for wind power clusters.

关 键 词：风电预测 风电集群 条件生成对抗网络 迁移学习 时间卷积网络 

分 类 号：TM74[电气工程—电力系统及自动化]