| 引用本文: | 蔡清淮,罗庆全,余涛,刘前进,刘熙鹏,潘振宁.基于主动迁移学习的负荷辨识泛化方法[J].电力自动化设备,2025,45(3): |
| CAI Qinghuai,LUO Qingquan,YU Tao,LIU Qianjin,LIU Xipeng,PAN Zhenning.Generalization method for load identification based on active transfer learning[J].Electric Power Automation Equipment,2025,45(3): |
|
| 摘要: |
| 为实现低成本地提升负荷辨识方法对新场景负荷样本的识别精度，提出一种基于主动迁移学习的负荷辨识泛化方法，利用极少量标签数据和无标签数据来高效提升方法的泛化性能。该方法利用异构模型的共识筛选高质量的伪标注样本，并对预训练模型进行更新；设计一种考虑模型分歧和样本多样性的主动学习策略来标注高价值样本，可在大幅降低样本标注成本的同时实现模型的高效迁移。在2个公开数据集中的实验对比，验证了所提方法的优越性。 |
| 关键词: 负荷辨识 泛化 主动学习 迁移学习 伪标注 异构模型 |
| DOI：10.16081/j.epae.202411013 |
| 分类号:TM714;TP18 |
| 基金项目:国家自然科学基金委员会-国家电网公司智能电网联合基金资助项目(U2066212)；广州市基础研究计划基础与应用基础研究项目(SL2022A04J01135)；广东省科技计划项目(广东省电网智能量测与先进计量企业重点实验室(2021年度))(2021B1212050014) |
|
| Generalization method for load identification based on active transfer learning |
|
CAI Qinghuai1, LUO Qingquan1, YU Tao1,2, LIU Qianjin1, LIU Xipeng1, PAN Zhenning1
|
|
1.School of Electric Power, South China University of Technology, Guangzhou 510640, China;2.Guangdong Provincial Key Laboratory of Intelligent Measurement and Advanced Metering of Power Grid, Guangzhou 510640, China
|
| Abstract: |
| In order to realize identification accuracy improvement of load identification methods to load samples in new scenarios with low cost, a load identification generalization method based on active transfer learning is proposed, which utilizes very few labelled data and unlabelled data to efficiently improve the generalization performance of the method. The method utilizes the consensus of heterogeneous models to select high-quality pseudo-labelled samples and updates the pre-trained model. An active learning strategy considering model disagreement and sample diversity is designed to label high-value samples, which can realize efficient model transfer while significantly reducing sample labeling cost. Experimental comparison of two public datasets verifies the superiority of the proposed method. |
| Key words: load identification generalization active learning transfer learning pseudo-labeling heterogeneous model |
