基于图信号处理的智能电表功率信号分解

智能电表的大规模部署,使得对电表采集的低频信号进行数据分析成为一个研究热点。以非侵入式负荷监测为背景,研究基于图信号处理(GSP)的低频功率信号分解算法。首先,将功率信号分解定义为最小化求解问题,并引入基于图转移矩阵的全局变化量作为正则项。然后,分两步对该优化问题求解:第1步最小化正则项得到满足图信号全局变化量最小的近似解;第2步以该解为基础,利用模拟退火算法对目标函数和约束条件迭代寻优。最后利用开源数据库REDD进行仿真,验证了该算法在分类准确率上的优势,且与其他算法相比对训练数据的依赖性较小。

Power Signal Disaggregation for Smart Meter Based on Graph Signal Processing

With the large-scale deployment of smart meters, data analysis based on low frequency signals collected by electric meter has become a research hotspot. Therefore, a low-rate power signal disaggregation algorithm based on graph signal processing(GSP)is studied in the background of non-intrusive load monitoring. Firstly, the power signal disaggregation problem is defined as a minimization problem, the total graph variation based on graph shift matrix is introduced as a regularization term. Then, the optimization problem is solved in two steps, which are minimizing the regularization term to find the approximate solution with minimum variation, using SA algorithm to minimize the objective function and constraint interactively based on the approximate solution. Finally, the open-access REDD dataset is used to demonstrate the advantage of the proposed algorithm in classification accuracy, especially for low power and multi-state load. Compared with other algorithms, the proposed algorithm is less dependent on training data.