摘要
针对目前的非侵入式负荷监测算法所需运算成本高、难以实用推广的现状,提出一种低运算成本的基于混合特征图的非侵入式负荷监测算法。首先,提取设备的功率特征和稳态电压-电流轨迹图特征,将设备功率特征进行维数变换后与电压-电流轨迹特征图组合,得到设备混合特征图。该特征图以小尺寸灰度图为载体,减小了硬件存储与模型算力的成本。然后,基于LeNet卷积神经网络建立设备辨识模型,以混合特征图为输入,实现对设备种类的辨识。最后,使用PLAID数据集对所提算法的结果准确性与计算性能进行测试。结果表明:所提算法的设备辨识准确率可达92.7%,与辨识准确率相差小于1%的同类算法相比,在算法参数量和运算量方面减少了99%,能有效减少NILM的运算成本。
In view of the current situation that non-intrusive load monitoring algorithm required high computation cost and was difficult to popularize,a low computation cost non-invasive load monitoring algorithm based on mixed feature graph was proposed.Firstly,the power features and steady-state U I track diagram features of the device were extracted,and the power features of the device were transformed by dimension and combined with the U I track feature diagram to obtain the mixed feature diagram of the device.The feature map wos supported by a small-scale gray map,which reduced the cost of hardware storage and model computing power.Then,the equipment identification model was established based on LeNet convolutional neural network,and the mixed feature map was input to realize the identification of equipment types.Finally,PLAID data set was used to test the accuracy and computational performance of the proposed algorithm.The results show that the equipment identification accuracy of the proposed algorithm can reach 92.7%.Compared with similar algorithms whose identification accuracy is less than 1%,the number of algorithm parameters and the amount of computation are reduced by 99%,and the computation cost of the NILM can be effectively reduced.
作者
梁浚杰
杨舒惠
鲍海波
莫江婷
李江伟
郭小璇
LIANG Junjie;YANG Shuhui;BAO Haibo;MO Jiangting;LI Jiangwei;GUO Xiaoxuan(Nanning Power Supply Bureau of Guangxi Power Grid Co.,Ltd.,Nanning 530031,China;Guangxi Key Laboratory of Power System Optimization and Energy Technology,Guangxi University,Nanning 530004,China;Electric Power Research Institute of Guangxi Power Grid Co.,Ltd.,Nanning 530023,China)
出处
《广西大学学报（自然科学版）》
CAS
北大核心
2023年第1期132-139,共8页
Journal of Guangxi University（Natural Science Edition）
基金
广西重点研发计划项目(桂科AB22080022)
中国南方电网公司科技项目(GXKJXM20190717)。
