电器与能效管理技术 ›› 2023, Vol. 0 ›› Issue (7): 55-60.doi: 10.16628/j.cnki.2095-8188.2023.07.009

• 识别与预测技术 • 上一篇    下一篇

基于多特征融合的非侵入式负荷辨识方法研究

魏星琦1, 张宸宇2, 缪惠宇2, 陈舒2   

  1. 1.江苏省电力有限公司, 江苏 南京 210003
    2.江苏省电力有限公司 电力科学研究院, 江苏 南京 211103
  • 收稿日期:2023-02-17 出版日期:2023-07-30 发布日期:2023-09-20
  • 作者简介:魏星琦(1990—),男,工程师,主要从事电力系统分析与控制研究。|张宸宇(1989—),男,高级工程师,博士,主要从事智能保护及变电站综合自动化研究。|缪惠宇(1992—),男,博士,主要从事电力电子在电力系统中的应用研究。
  • 基金资助:
    2020年江苏省电力有限公司科技项目(J2022021)

Research on Non-Intrusive Load Identification Method Based on Multi-Feature Fusion

WEI Xinqi1, ZHANG Chenyu2, MIAO Huiyu2, CHEN Shu2   

  1. 1. Jiangsu Electric Power Co.,Ltd., Nanjing 210003, China
    2. Electric Power Research Institute of Jiangsu Electric Power Co.,Ltd, Nanjing, 211103, China
  • Received:2023-02-17 Online:2023-07-30 Published:2023-09-20

摘要:

针对传统的非侵入式负荷辨识算法在低频采样数据上辨识准确率较低的问题,提出了一种基于格拉姆角场(GAF)和特征融合的非侵入式负荷辨识算法。通过GAF编码将功率信息转换成彩色图像特征,从而提升辨识度。将图像和功率数据分别输入到卷积神经网络和反向传播神经网络中提取特征,实现特征融合,并作为新的特征参与辨识。用公开数据集对所提辨识方法进行多方面的验证,并与其他分类算法进行对比。结果表明,图像特征所携带的信息量更多,加强了特征的代表性,特征融合能够解决图像编码过程中的信息丢失问题,从而提高了模型的负荷辨识能力。

关键词: 非侵入式负荷辨识, 格拉姆角场, 特征融合, 深度学习

Abstract:

Aiming at the problem that the accuracy of traditional non-intrusive load identification algorithms is low in low-frequency sampling data,a non-intrusive load identification algorithm based on the gram angle field (GAF) and feature fusion is proposed.The collected power information is converted into color image features by GAF technology to improve the identification.The image and power data are input into the convolutional neural network and the backpropagation neural network respectively for feature extraction to achieve feature fusion,which is used as the new feature for identification.The identification method is verified in many aspects of the public data set and compared with different classification algorithms.The results show that the image features carry more information,strengthen the representation of features,and feature fusion can solve the problem of information loss in the process of image coding,and improve the load identification ability of the model.

Key words: non-intrusive load identification, gramian angular field (GAF), feature fusion, deep learning

中图分类号: