基于改进条件生成对抗网络的需求响应潜力预测
Demand Response Potential Prediction Based on Improved Conditional Generative Adversarial Networks
房子衿 1周波 2林顺富 2赵建立 3吴丹 3许亮峰1
作者信息
- 1. 上海电力大学,上海 200090
- 2. 上海电力大学,上海 200090;上海市智能电网需求响应技术重点实验室,上海 200063
- 3. 国网上海市电力公司,上海 200120
- 折叠
摘要
为了解决现行各类基线负荷预测方法对时序性的依赖及多影响因素分类分析评估不准确等问题,提出了一种基于改进条件生成对抗网络(CGAN)和图卷积神经网络(GCN)的需求响应潜力预测方法.首先,引入融合CGAN和Wasserstein生成对抗网络(WGAN)的Wasserstein条件生成对抗网络(WCGAN),并利用历史数据训练生成器和判别器,估计基线负荷;然后,充分考虑基线与响应负荷的不确定性,提出了一种基于WCGAN和GCN的需求响应负荷预测方法;最后,采用实际负荷数据对所提方法的有效性进行了分析.
Abstract
In order to solve the problems such as the dependence of the existing various baseline load estimation methods on temporality and the inaccurate assessment of the classification analysis of multiple influencing factors,a user-side flexibility resource response potential assessment method based on the improved conditional generative adversarial networks(CGAN)and graph convolutional neural networks(GCN)is proposed.Firstly,a fusion of wasserstein generative adversarial networks(WGAN)and CGAN is introduced to create.Wasserstein conditional generative adversarial networks(WCGAN),and historical data are used to train the generator and discriminator to estimate the baseline load;then,with full consideration of the uncertainty of the baseline and response loads,we propose an approach based on the WCGAN and GCN;finally,the effectiveness of the proposed method is analysed by using actual load data.
关键词
需求响应/基线负荷/响应潜力/Wasserstein条件生成对抗网络/图卷积神经网络Key words
demand response/baseline load/response potential/Wasserstein conditional generative adversarial network/graph convolutional neural network引用本文复制引用
基金项目
国家自然科学基金(51977127)
上海市教育发展基金会和上海市教育委员会"曙光计划"(20SG52)
出版年
2024
NETL
NSTL
万方数据