摘要
提出一种基于遗传算法和双向长短时记忆网络的非侵入式负荷分解方法,通过卷积神经网络对总能耗波形进行特征提取,借助双向长短时记忆网络的双向传播特性对能耗波形特征进行训练,引入遗传算法对网络的超参数进行优化,以提高网络性能。利用UK-DALE数据集对所提方法进行对比验证,证明了本文方法对建筑分项能耗预测的有效性和准确性,以及在波动能耗跟踪上的优良性能。
A non-intrusiveload disaggregation method based on genetic algorithm and bidirectional longshort-termnmemory network is proposed.The feature extraction of total energy consumption waveform is carried out by use of convolutional neural network,and the energy consumption waveform featuresare trainedaccording to the bidirectional propagation characteristics of bidirectional long short-term memory(LSTM)network;genetic algorithm is introduced to optimize the hyperparameters of the network,so as to improve network performance.The comparison and validation are carried out for the proposed method by use of the UK-DALE dataset,and thus the effectiveness and accuracy of the proposed method in predicting the energy consumptionofbuilding sub-items,as well as their excellent performancein tracking fluctuating energy consumption are demonstrated.
作者
郑宇
杨祝涛
周强
ZHENG Yu;YANG Zhutao;ZHOU Qiang(China Southwest Architectural Design and Research Institute Corp.Ltd,Chengdu 610042,China)
出处
《建筑电气》
2023年第11期23-29,共7页
Building Electricity
关键词
建筑能耗
非侵入式负荷分解
卷积神经网络
长短时记忆网络
遗传算法
深度学习
特征提取
分项能耗预测
building energy consumption
non-intrusive load disaggregation
convolutional neural network
LSTM network
genetic algorithm
in-depth learning
feature extraction
itemized energy consumption prediction