基于生成对抗网络的风光出力场景生成  

Wind-Photovoltaic Power Generation Scenario Generation Based on GAN

作　　者：郭子铭 GUO Ziming(North China Electric Power University,Beijing 102206,China)

机构地区：[1]华北电力大学,北京102206

出　　处：《工业技术与职业教育》2025年第2期8-12,18,共6页Industrial Technology and Vocational Education

摘　　要：新型电力系统运行与规划中涉及大量新能源场景的建模和分析,场景选取的合理性对系统运行和规划的计算效率和可靠性有重要影响。为此,提出基于深度学习的VAE-DCGAN模型,该模型使用变分自动编码器作为生成对抗网络的生成器部分,使其学习历史数据的出力特征,并通过与判别器的博弈训练生成大量风光出力场景。利用累积概率分布和最大均值差异距离等指标,对生成场景的优劣进行评估。结果表明,VAE-DCGAN模型的生成数据和真实数据的MMD距离为0.0329,模型可以较好地学习到风光历史数据的出力特征。The operation and planning of the new power system involve a large number of modeling and analysis of new energy scenarios,and the rationality of the scenarios has a significant impact on the computational efficiency and reliability of system operation and planning.To this end,this study proposes a scenario generation method for wind and solar generation based on the VAE-DCGAN model,which uses a variational autoencoder as the generator part of the generative adversarial network,learns the power generation characteristics of data,and generates a large number of wind and solar power generation scenarios through a game training with the discriminator.The superiority of the generated scenarios is evaluated using metrics such the difference distance of the cumulative probability distribution and the maximum mean.The results show that the MMD distance between the generated data and the real data of the VAEDCGAN model is 0.0329,and the model can better learn the power generation characteristics of historical wind and solar data.

关 键 词：场景生成 变分自动编码器 生成对抗网络 新能源出力 最大均值差异距离 

分 类 号：TM715[电气工程—电力系统及自动化]