| 引用本文: | 冯昌森,刘攀,王佳颖,文福拴,张有兵.利用有限低频信息的居民用户非侵入负荷监测算法[J].电力自动化设备,2023,43(11):181-187 |
| FENG Changsen,LIU Pan,WANG Jiaying,WEN Fushuan,ZHANG Youbing.Non-intrusive load monitoring algorithm of residential users using limited low-frequency information[J].Electric Power Automation Equipment,2023,43(11):181-187 |
|
| 摘要: |
| 通过更多信息特征或高频采样技术提高识别准确率的负荷监测算法，会增加信息采样阶段的成本和边缘数据处理的难度，提出一种基于有限低频信息的非侵入式负荷监测算法。设计最佳事件检测器，该检测器根据滑动窗口采集聚合负荷数据，并根据统计特征指标判断电器投切位置；将事件发生前后的功率序列作为识别特征，利用互补集合经验模态分解算法分解出功率序列中的多阶本征模态函数和最终趋势，绘制分解结果的二维图像并将其输入卷积神经网络进行训练和识别，从而实现仅基于有限低频采样信息就可高精确率地识别负荷。基于公开数据集的仿真结果验证了所提算法的有效性。 |
| 关键词: 非侵入式负荷监测 事件检测 互补集合经验模态分解 卷积神经网络 负荷识别 |
| DOI：10.16081/j.epae.202304014 |
| 分类号:TM715 |
| 基金项目:国家自然科学基金资助项目(52107129, U22B20116) |
|
| Non-intrusive load monitoring algorithm of residential users using limited low-frequency information |
|
FENG Changsen1, LIU Pan1, WANG Jiaying2, WEN Fushuan3,4, ZHANG Youbing1
|
|
1.College of Information Engineering, Zhejiang University of Technology, Hangzhou 310023, China;2.Marketing Service Center of State Grid Zhejiang Electric Power Co.,Ltd.,Hangzhou 311121, China;3.Hainan Institute, Zhejiang University, Sanya 572000, China;4.College of Electrical Engineering, Zhejiang University, Hangzhou 310027, China
|
| Abstract: |
| The load monitoring algorithm which uses more information features or high-frequency sampling technology to improve the identification accuracy will increase the cost of information sampling stage and the difficulty of edge data processing, a non-intrusive load monitoring algorithm based on limited low-frequency information is proposed. An optimal event detector is designed, which collects aggregated load data according to a sliding window and judges the switching position of electrical appliances according to the statistical characteristic index. The power sequences before and after event occurrence are taken as the identification features, and the complementary ensemble empirical mode decomposition algorithm is used to decompose the power sequences into multi-order intrinsic mode functions and a final trend, a two-dimensional image of the decomposition results is drawn and input into the convolution neural network for training and identification, thus high-accuracy load identification is realized by only using limited low-frequency sampling information. The simulative results based on public dataset verify the effectiveness of the proposed algorithm. |
| Key words: non-intrusive load monitoring event detection complementary ensemble empirical mode decomposition convolutional neural network load identification |
