基于改进型颜色编码与密集连接网络的非侵入式负荷监测  

Non-invasive load monitoring based on improved color coding and dense convolutional network

作　　者：丁健 陈鉴祥 刘旺朋 丁一帆 王谱宇 DING Jian;CHEN Jianxiang;LIU Wangpeng;DING Yifan;WANG Puyu(School of Automation,Nanjing University of Science and Technology,Nanjing 210094,China)

机构地区：[1]南京理工大学自动化学院,南京210094

出　　处：《浙江电力》2025年第2期3-12,共10页Zhejiang Electric Power

基　　金：江苏省自然科学基金(BK20242048)。

摘　　要：现有基于V-I轨迹的非侵入式负荷监测研究,存在对于轨迹相似的电器设备识别准确率较低、网络较复杂的问题。针对该问题提出了一种基于改进型颜色编码与密集连接网络的非侵入式负荷监测方法。首先,分析了基于滑动窗的事件检测算法的原理;其次,提出了一种改进型颜色编码V-I轨迹的绘制方法,以解决小功率区间电器种类繁多的问题,提高V-I轨迹的区分度;然后,将所绘制的V-I轨迹用于自建密集连接网络的训练,得到了一个适用于低分辨率V-I轨迹的识别模型;最后,在PLAID与WHITED数据集上进行验证,结果表明:所提方法具有更高的准确率和识别效率。Current research on non-intrusive load monitoring(NILM)based on V-I trajectories faces challenges such as low recognition accuracy for appliances with similar trajectories and complex network structures.To address these issues,the paper proposes a non-intrusive load monitoring method based on improved color coding and dense convolutional network(DenseNet).First,the paper analyzes the principles of the event detection algorithm based on a sliding window.Then,it introduces an improved color coding algorithm for plotting V-I trajectories to address the problem of diverse appliance types in low-power ranges and to enhance the differentiation of V-I trajectories.Next,the plotted V-I trajectories are used to train the dense convolutional network,resulting in a recognition model suitable for low-resolution V-I trajectories.Finally,validation on the PLAID and WHITED demonstrates that the proposed method achieves higher accuracy and recognition efficiency.

关 键 词：非侵入式负荷监测 V-I轨迹 图像识别 密集连接网络 家用电器 

分 类 号：TM714[电气工程—电力系统及自动化] TP391.41[自动化与计算机技术—计算机应用技术]