摘要
为了解决传统商业建筑负荷分解算法存在的精度低、稳定性差等问题,提出了基于注意力机制-长短期记忆(Attention-LSTM)人工神经网络的负荷分解模型,该模型能充分学习负荷的内在特征,并采用Attention机制对影响负荷分解的因素赋予权值,突出重点因素。以上海某商业建筑的低频数据集为实验分析数据,进行十折交叉试验,结果表明所提模型分解精度优于其他模型,具有较好的泛化能力和较强的稳定性。
Traditional energy disaggregation algorithm has the problems of low accuracy and poor stability.Therefore,energy disaggregation model is proposed based on attention-LSTM network,which can fully learn the inherent characteristics of load and assign weights to factors affecting load decomposition by using the Attention mechanism.Low-frequency data sets of commercial buildings in Shanghai are taken as experimental data,a ten-fold crossover test is performed.The results show that this model is better than other models in decomposition accuracy,has better stability and generalization ability.
作者
杨秀
曹泽棋
张美霞
胡钟毓
李莉华
YANG Xiu;CAO Zeqi;ZHANG Meixia;HU Zhongyu;LI Lihua(School of Electrical Engineering,Shanghai University of Electric Power,Shanghai 200082,China;State Grid Shanghai Electrical Power Research Institute,Shanghai 200080,China)
出处
《智慧电力》
北大核心
2020年第9期89-95,共7页
Smart Power
基金
国家自然科学基金资助项目(51807114)
上海市科委资助项目(18DZ1203200)。
关键词
商业建筑
负荷分解
深度学习
长短期记忆网络
注意力机制
commercial buildings
energy disaggregation
deep learning
long short-term memory network
attention mechanism
