摘要
非侵入式负荷分解可将家庭总电表监测的数据分解为单台用电器的状态、能耗等信息。针对目前神经网络模型对长激活且周期运行的电器分解效果较差的缺点,本研究提出1种基于自注意力的非侵入式负荷分解模型(Attention-BO模型)。Attention-BO模型以稳态有功功率变化为负荷特征,利用卷积神经网络进行特征提取,通过自注意力增强对输入序列中远距离相关性的建模,获得与各点最相关的特征。之后,使用贝叶斯优化对模型中的超参数进行调节以提高精度。最后,在UK-dale数据集上对模型进行训练和测试,实验结果表明模型具有较小的误差,并且在新环境下的泛化性能较好。
Non-intrusive load disaggregation can disaggregate the data monitored by the household master meter into the state and energy consumption of individual appliances.To address the poor efficiency of the existing neural network models in disaggregation of appliances with long-term activation and periodic operation,this study proposed a non-intrusive load disaggregation model based on self-attention(Attention-BO model).The proposed model takes the steady-state active power variation as the load feature,utilizes convolutional neural networks for feature extraction,and enhances the modeling of long-range correlation in the input sequence by self-attention to obtain the most relevant features with each point.Then,Bayesian optimization is used to adjust the hyperparameter in the model to improve the accuracy.After that,the model is trained and tested on the UK-dale dataset,and the results show that the model exhibits lower error and better generalization in the new environment.
作者
赵安军
崔朴方
于军琪
赵啸
陈一仁
ZHAO Anjun;CUI Pufang;YU Junqi;ZHAO Xiao;CHEN Yiren(School of Building Services Science and Engineering,Xi'an University of Architecture and Technology,Xian 710055,China;State Power Investment Group Shaanxi New Energy Co,Xian 710061,China)
出处
《建筑科学》
CSCD
北大核心
2023年第4期113-121,212,共10页
Building Science
基金
国家自然科学基金重点项目“含氢多能源供需系统协同运行的基础理论与关键技术”(62192750)。
关键词
负荷分解
自注意力机制
卷积神经网络
非侵入式负荷监测
load disaggregation
self-attention mechanism
convolutional neural network
non-intrusive load monitoring
