摘要
现有的非侵入式负荷分解算法往往需要大量电器设备级的负荷数据才能保证分解精度,但由于用户对隐私性的考虑以及安装成本过高等问题,很难获取这些数据。因此,构建一种能深度挖掘电力负荷数据时序特性和电器相关性的时序生成对抗网络。利用降维网络对所有电器有功功率序列组成的高维向量进行降维以降低计算的复杂度,通过复原网络将结果还原为高维向量。基于电器运行状态和深度学习的非侵入式分解方法,运用卷积神经网络-双向门控循环单元构建状态复杂电器的负荷分解回归模型,对状态简单电器利用深度神经网络构建负荷识别分类模型。通过对比其他数据生成方法,以及改变典型公开数据集中生成数据比例所得的负荷分解结果验证了所提方法的有效性。
The existing non-intrusive load decomposition algorithms often require a large number of electrical load data at equipment level to ensure the decomposition accuracy,but it is difficult to obtain these data due to the user's consideration of privacy and high installation cost.Therefore,a time-series generative adversarial network which can deeply explore the time-series characteristics of power load data and the correlation of electrical appliances is constructed.The dimensionality reduction network is used to reduce the dimensionality of the high-dimensional vector composed of the active power sequence of all electrical appliances to reduce the computational complexity,and the result is restored to the high-dimensional vector through the restoration network.Based on the non-intrusive decomposition method of electrical appliance operation status and deep learning,the load decomposition regression model of complex-state electrical appliances is built by using the convolutional neural network-bidirectional gated recurrent unit,and the load identification and classification model of simple-state electrical appliances is constructed by using the deep neural network.The effectiveness of the proposed method is verified by comparing other data generation methods and the load decomposition results obtained by changing the proportion of the generated data in typical public datasets.
作者
罗平
朱振宇
樊星驰
孙博宇
张帆
吕强
LUO Ping;ZHU Zhenyu;FAN Xingchi;SUN Boyu;ZHANG Fan;LYU Qiang(School of Automation,Hangzhou Dianzi University,Hangzhou 310018,China;School of ITMO Joint Institute,Hangzhou Dianzi University,Hangzhou 310018,China;School of Information Engineering,Hangzhou Dianzi University,Hangzhou 310018,China)
出处
《电力系统自动化》
EI
CSCD
北大核心
2024年第2期71-81,共11页
Automation of Electric Power Systems
基金
浙江省自然科学基金资助项目(LY20E070004)
国家自然科学基金资助项目(62073108)。
关键词
非侵入式负荷分解
对抗生成网络
降维网络
卷积神经网络-双向门控循环单元
深度神经网络
non-intrusive load decomposition
generative adversarial network
dimensionality reduction network
convolutional neural network-bidirectional gated recurrent unit
deep neural network
