基于可解释性条件生成对抗网络的台风气象负荷场景生成方法  

Scenario Generation Method of Typhoon Meteorological Load Based on Interpretability Conditional Generative Adversarial Network

作　　者：罗萍萍 盛奥 林济铿[2] 马骞[3] 许琴 刘一鸣 LUO Pingping;SHENG Ao;LIN Jikeng;MA Qian;XU Qin;LIU Yiming(School of Electric Power Engineering,Shanghai University of Electric Power,Shanghai 200090,China;Electronic and Information Engineering,Tongji University,Shanghai 201804,China;China Southern Power Grid Company Limited,Guangzhou 510663,China;Guangdong Electric Power Design Institute Co.,Ltd.of China Energy Engineering Group,Guangzhou 510663,China)

机构地区：[1]上海电力大学电气工程学院,上海市200090 [2]同济大学电子与信息工程学院,上海市201804 [3]中国南方电网有限责任公司,广东省广州市510663 [4]中国能源建设集团广东省电力设计研究院有限公司,广东省广州市510663

出　　处：《电力系统自动化》2025年第2期186-197,共12页Automation of Electric Power Systems

基　　金：国家自然科学基金资助项目(51177107)。

摘　　要：台风气象下电网负荷将会出现剧烈波动且威胁到电网安全稳定运行,亟需一种有效的方法来生成相应的负荷需求场景。文中提出一种面向稀少历史样本、基于可解释性条件生成对抗网络(CGAN)的台风负荷场景生成方法。首先,对历史台风负荷进行修正,并根据台风登陆位置、等级等信息对其进行标签分类。然后,提出一种两阶段数据扩充策略以应对数据匮乏问题,第1阶段利用历史台风日负荷序列之间的横纵向相关性信息进行样本扩充,第2阶段利用台风日与非台风日负荷之间的残差信息进一步进行样本扩充。最后,提出基于特征影响指标的CGAN因果解释方法,刻画了不同特征对于模型结果的调控力度大小。算例证实了文中所提模型及方法的有效性和先进性。Under typhoon meteorology,the load of the power grid will experience severe fluctuations,posing a threat to the safe and stable operation of the power grid.An effective method is urgently needed to generate corresponding load demand scenarios.A typhoon load scenario generation method is proposed based on interpretability conditional generative adversarial network(CGAN)for rare historical samples.First,the historical typhoon load is adjusted and classified based on the information such as typhoon landfall locations and levels.Then,a two-stage data augmentation strategy is proposed to address the issue of data scarcity.The first stage utilizes the horizontal and vertical correlation information between historical typhoon daily load sequences for sample augmentation,while the second stage utilizes the residual information between typhoon and non-typhoon daily loads for further sample augmentation.Finally,a CGAN causal interpretation method based on feature influence indicators is proposed,which characterizes the regulation degree of different features on the model results.Numerical cases show that the proposed model and method are effective and progressive.

关 键 词：台风气象 人工智能 负荷需求 场景生成 可解释性 条件生成对抗网络 

分 类 号：TM714[电气工程—电力系统及自动化] P444[天文地球—大气科学及气象学]