一种基于CNN-LSTM的改进CGAN光伏短期出力场景生成方法  

IMPROVED CGAN PHOTOVOLTAIC SHORT-TERM OUTPUT SCENARIO GENERATION METHOD BASED ON CNN-LSTM

作　　者：秦卫民 唐昊[1] 任曼曼 梁肖 王涛[1] 陈韬 Qin Weimin;Tang Hao;Ren Manman;Liang Xiao;Wang Tao;Chen Tao(School of Electrical Engineering and Automation,Hefei University of Technology,Hefei 230009,China;State Grid Anhui Electric Power Company,Hefei 230061,China;School of Physics and Electronic Engineering,Fuyang Normal University,Fuyang 236041,China)

机构地区：[1]合肥工业大学电气与自动化工程学院,合肥230009 [2]国网安徽省电力有限公司,合肥230061 [3]阜阳师范大学物理与电子工程学院,阜阳236041

出　　处：《太阳能学报》2025年第4期263-272,共10页Acta Energiae Solaris Sinica

基　　金：国家自然科学基金(62273130);安徽省自然科学基金(2108085UD01);安徽省高校协同创新项目(GXXT-2023-032);企业委托项目(JYDW-231273)。

摘　　要：该文考虑新能源机组出力数据的时空特征,设计一种带有卷积神经网络和长短期记忆网络的判别器网络结构,并使用推土机(EM)距离作为判别器的损失函数,提出一种基于条件对抗生成网络的新能源短期场景生成方法。该方法让模型中的判别器与生成器进行对抗并不断优化,使生成器网络更加准确地提取到条件值及噪声分布与样本分布之间的映射关系,从而更好地生成新能源机组出力场景。该文使用开源的光伏出力数据对模型进行验证和测试,相对于传统的生成式对抗网络方法,所提模型能更加准确地生成契合历史数据特征的新能源出力场景集。This article considers the spatiotemporal characteristics of output data of new energy units and designs a discriminator network structure with convolutional neural networks and long short-term memory networks.The EM distance is used as the loss function of the discriminator,and a new energy short-term scene generation method based on conditional adversarial generation network is proposed.This method trains the generator network through game theory between the generator and discriminator in the model,enabling it to more accurately extract the mapping relationship between conditional values,noise distribution,and sample distribution,thereby better generating new energy unit output scenarios.This article uses open-source photovoltaic output data to validate and test the model,and compares it with the generative adversarial network scene generation method based on fully connected structures.The results show that the proposed model can more accurately generate short-term output scene sets that match the characteristics of real samples.

关 键 词：光伏发电 场景生成 生成式对抗网络 长短期记忆网络 卷积神经网络 不确定性 

分 类 号：TM615[电气工程—电力系统及自动化] TP183[自动化与计算机技术—控制理论与控制工程]