基于条件深度卷积生成对抗网络的新能源发电场景数据迁移方法

Renewable Power Generation Data Transferring Based on Conditional Deep Convolutions Generative Adversarial Network

针对在历史数据缺失的情况下,现有的新能源发电场景生成方法存在精度较低甚至失效的问题,提出一种基于条件深度卷积生成对抗网络(conditional deep convolutions generative adversarial network,C-DCGAN)的新能源发电场景数据迁移方法。该方法以历史数据大规模缺失的新能源电站为目标电站,以历史数据完整的邻近新能源电站为源电站,通过生成对抗网络模型学习源电站与目标电站之间的场景数据映射关系,进而根据源电站场景数据,生成目标电站场景数据,且所生成的数据符合真实场景数据分布规律。采用实际风电数据集对所提算法和模型进行验证,并应用若干统计学指标,分别对文中模型与条件生成对抗网络(conditional generative adversarial network,CGAN)模型所迁移生成的数据进行对比评估,结果表明所提算法与模型能够更加准确地生成新能源发电场景数据。

When the historical renewable power generation data are missing,the data-driven scenario generation methods may be invalid.To deal with this issue,in this paper,we propose a conditional deep convolutions generative adversarial network(C-DCGAN)to recover and transfer the historical renewable power generation data.First,a renewable power plant with a lot of missing historical data is considered as the target plant,and a neighboring plant with sufficient historical data is considered as the source plant.Then,the proposed C-DCGAN model learns the correlation between the data in the target plant and the source plant.After that,the C-DCGAN model recovers the missing data in the target plant based on the historical data in the source plant.In this way,the historical data in the source plant is transferred to the target plant.Finally,the numerical experiments have been carried out based on a wind farm data,and some statistical indicators are used to evaluate the data recovered by the proposed C-DCGAN model and the conditional generative adversarial network(CGAN)model.The simulation results show that the proposed C-DCGAN model has better performance in renewable power generation data transferring compared with the CGAN model.