基于对抗神经网络和K-means++的配电网光伏接入能力评估方法  

Photovoltaic access capability evaluation of distribution networks based on adversarial neural networks and K-means++clustering

作　　者：张家美 孙凯[1] 孙沛 ZHANG Jiamei;SUN Kai;SUN Pei(State Key Laboratory of Power System Operation and Control,Tsinghua University,Beijing 100084,China;Pingliang Power Supply Company of State Grid Gansu Electric Power Company,Pingliang 744021,China)

机构地区：[1]清华大学新型电力系统运行与控制全国重点实验室,北京100084 [2]国网甘肃省电力公司平凉供电公司,甘肃平凉744021

出　　处：《电工电能新技术》2025年第2期65-73,共9页Advanced Technology of Electrical Engineering and Energy

基　　金：国家电网有限公司总部管理科技项目(5400-202233168A-1-1-ZN)。

摘　　要：分布式光伏的大量接入对配电网的规划与运行产生了重大影响,如何评估现有配电网的光伏最大可接入能力是指导分布式光伏安全消纳的基础。考虑光伏出力不确性对评估结果的影响,本文提出基于生成式对抗网络(GAN)和K-means++聚类的配电网光伏可接入容量评估方法。该方法通过GAN的生成器和判别器之间的博弈,学习光伏时序数据的分布特征,并产生足量可以反映真实天气情况的伪造光伏出力场景。通过K-means++聚类方法进行场景削减与典型场景生成。基于典型场景,在满足安全运行约束的情况下构建以光伏接入容量最大为目标的优化模型。运用二阶锥松弛将非线性的配电网潮流模型转化为凸优化模型。最后,基于IEEE 33节点配电网对所提出方法进行测试,并与机会约束优化、鲁棒优化等随机优化方法进行对比。结果验证了本文所提的优化评估方法可以综合考虑不同天气类型的影响,提供更符合实际运行要求的光伏接入方案。The widespread and extensive access of distributed photovoltaics has a significant impact on the planning and operation of distribution networks.How to evaluate the maximum accessible capacity of existing distribution networks is the foundation for guiding the safe consumption of distributed photovoltaics.Considering the impact of photovoltaic output uncertainty on the evaluation results,this paper proposes a method for evaluating the photovoltaic accessible capacity of distribution networks based on generative adversarial networks(GAN)and K-means++clustering analysis.This method automatically learns the complex spatiotemporal relationships explicitly modeled between photovoltaic output data through GAN,generating a sufficient amount of“false”data that can reflect real weather conditions.Using K-means++clustering analysis method for scene reduction and typical scene generation,based on typical scenarios,an optimization model is constructed with the goal of maximizing photovoltaic access capacity while meeting the constraints of safe operation and using second-order cone optimization to linearize the power flow model of the distribution network.Finally,the results of the proposed method are analyzed and evaluated based on the IEEE 33 node distribution network.The calculation results verify that the capacity optimization method proposed in this article can comprehensively consider the impact of different weather types and provide a photovoltaic access scheme that is more in line with actual operational requirements.

关 键 词：生成式对抗网络 K-means++ 分布式光伏 容量评估 典型场景 

分 类 号：TM732[电气工程—电力系统及自动化]