摘要
针对目前负荷监测领域存在模型分解准确率低、训练周期长、泛化性能差的问题,文章构建了基于通道注意力机制和双向门控循环单元的非侵入式负荷监测模型,利用搭建的序列到点编码–解码结构,将智能电表入口处总功率序列与待测目标设备序列中点在模型上进行映射训练。使用功率嵌入层对负荷序列的输入过程进行优化,将离散的负荷总功率序列通过功率嵌入矩阵映射到高维紧密向量空间;采用一维卷积神经网络提取负荷序列的局部特征,双向门控循环单元提取负荷的长序列依赖关系,同时融合通道注意力机制强化对目标设备重要信号特征的学习,挖掘目标设备状态与负荷功率之间的关联。在基于能量分解模型基准框架下,利用公开数据集REDD和UK-DALE进行实验,与现有两种典型负荷分解模型进行比较分析,实验结果表明,文章构建的模型在减少网络训练时间和参数的前提下,有效检测了目标设备的开关状态,显著提升了负荷分解准确性。
A non-intrusive load monitoring model based on the channel attention mechanism and BIGRU is constructed to address the current problems of low accuracy of model decomposition,long training period and poor generalization performance in the field of load monitoring.Using the constructed sequence-to-point encoding-decoding structure,the total power sequence at the entrance of the smart meter is mapped to the midpoint of the sequence of target devices to be tested on the model for training.The input process of the load sequence is first optimized using a power embedding layer to map the discrete load total power sequence to a high-dimensional compact vector space via a power embedding matrix.A one-dimensional convolutional neural network is used to extract the local features of the load sequence,a bi-directional gated recurrent network(BIGRU)to extract the long sequence dependencies of the load,and a fused channel attention mechanism(Attention)to enhance the target device state and load power correlations are explored by incorporating a channel attention mechanism(Attention)to enhance the learning of important signal features of the target device.In the benchmark framework based on the energy decomposition model,experiments are conducted using the publicly available datasets REDD and UK-DALE to compare and analyse with two existing typical load decomposition models.The experimental results show that the model constructed in this paper effectively detects the switching status of the target device and significantly improves the load decomposition accuracy with reduced network training time and parameters.
作者
钱玉军
包永强
姜丹琪
张旭旭
雷家浩
QIAN Yujun;BAO Yongqiang;JIANG Danqi;ZHANG Xuxu;LEI Jiahao(School of Electric Power Engineering,Nanjing Institute of Technology,Nanjing 211167,Jiangsu Province,China;School of Information and Communication Engineering,Nanjing Institute of Technology,Nanjing 211167,Jiangsu Province,China)
出处
《电力信息与通信技术》
2023年第7期1-10,共10页
Electric Power Information and Communication Technology
基金
国家自然科学基金项目(62171217)。
作者简介
通信作者:钱玉军(1995),男,硕士研究生,研究方向为基于大数据框架下的非侵入负荷监测,577600985@qq.com;包永强(1973),男,教授,研究方向为信号处理、非侵入式负荷监测;姜丹琪(1999),女,硕士研究生,研究方向为大数据处理和非侵入式负荷检测;张旭旭(1997),男,硕士研究生,研究方向为信号处理和非侵入式负荷检测;雷家浩(1999),男,硕士研究生,研究方向为大数据处理和非侵入式负荷检测。
