谭嘉¹, 李知艺¹(), 杨欢¹, 赵荣祥¹, 鞠平^1,2

1.浙江大学 电气工程学院,杭州 310007
2.河海大学 能源与电气学院,南京 211100

收稿日期:2021-07-30出版日期:2021-12-28发布日期:2021-12-30
通讯作者:李知艺E-mail:zhiyi@zju.edu.cn
作者简介:谭 嘉(1997-),女,四川省达州市人,硕士生,从事人工智能在电力系统中的应用研究.
基金资助:国家自然科学基金(U2066601);国家自然科学基金(52007164)

A Multi-Level Collaborative Load Forecasting Method for Distribution Networks Based on Distributed Optimization

TAN Jia¹, LI Zhiyi¹(), YANG Huan¹, ZHAO Rongxiang¹, JU Ping^1,2

1. College of Electrical Engineering, Zhejiang University, Hangzhou 310007, China
2. College of Energy and Electrical Engineering, Hohai University, Nanjing 211100, China

Received:2021-07-30Online:2021-12-28Published:2021-12-30
Contact:LI Zhiyi E-mail:zhiyi@zju.edu.cn

摘要/Abstract

摘要： 当前,分布式新能源、电动汽车等新元素在配电网中涌现,改变了负荷的构成,丰富了负荷的内涵,给负荷预测带来了严峻挑战.事实上,用电负荷在配电网的多个电压层级以自下而上的方式聚合,但现有预测研究鲜少考虑此类层级化特征.为保障负荷自下而上的聚合一致性并且联合提升各层级的负荷预测性能,提出了一种基于分布式优化算法的用电负荷多层协同预测方法.首先,采用基于交替方向乘子法的分布式优化理念,构建了适配配电网层级特征、数据交互量少的多层协同预测框架.随后,提出了基于长短期记忆神经网络和联邦学习的具体预测方法,通过将底层负荷预测结果逐级聚合,能实现自下而上的配电网负荷一体化预测.算例结果表明,所提方法得到的用电负荷预测准确度高,应用前景好.
关键词: 多层负荷预测, 交替方向乘子法, 长短期记忆神经网络, 联邦学习
Abstract: At present, new elements such as distributed new energy and electric vehicles have emerged in the distribution network, which changes the composition of loads, enriches the connotation of loads, and poses severe challenges to load forecasting. In fact, loads are aggregated in a bottom-up manner in multiple voltage levels of the distribution network, but such hierarchical characteristics are rarely considered in current load forecasting researches. Therefore, a multi-level load collaborative forecasting method based on the distributed optimization algorithm is proposed aimed at ensuring the bottom-up aggregation consistency of loads and jointly improving the performance of load forecasting at all levels. First, the distributed optimization concept based on the alternating direction method of multipliers is adopted to construct a multi-level load collaborative forecasting framework which adapts to the hierarchical characteristics of distribution network and has less data interaction. Then, a specific forecasting method based on the long short term-memory neural network and federated learning is proposed. By aggregating the bottom load forecasting results step by step, the bottom-up integrated load forecasting of distribution network can be realized. The results of calculation examples show that the proposed method has a high accuracy and a great application prospect.
Key words: multi-level load forecasting, alternating direction method of multipliers, long short-term memory neural network, federated learning


PDF全文下载地址:

点我下载PDF