基于CFSFDP图拉普拉斯算法的非侵入式负荷监测方法

非侵入式负荷监测(NILM)是我国未来电网建设的重要发展方向之一。为克服传统非侵入式负荷监测方法的计算数据量大、辨识准确率较低等问题，提出了一种基于CFSFDP(快速密度峰值搜索算法)图拉普拉斯算法的非侵入式负荷监测方法。首先，该方法利用输入的设备有功功率数据采取快速密度峰值搜索聚类算法构建家用电器的功率阈值向量和先验图结构；然后结合图信号的平滑度特征和总功率信号构建图拉普拉斯二次型最优函数，利用Tikhonov正则化方法以迭代的方式求得最优解，从而实现用电负荷图信号的重构；最后根据功率阈值向量将图信号转换为功率信号，即可实现用户的非侵入式负荷监测。对某一家庭两天的实测用电数据进行仿真分析，得到如下结果：1)该方法对第一天的负荷辨识精度达到了100%，各用电设备消耗用电量比例与实际耗电量比例误差均低于3%。2) 该方法对第二天的负荷识别准确率达到了90.1%，相比于对比算法至少高了0.8%。对单个用电设备分解精度达到91%以上，设备的用电量误差不超过5%且低于对比算法。3) 当数据采样间隔增大为2min，所提算法的准确率、辨识精度和单设备分解精度都有所降低，但数值上优于对比算法，并且有更优的时间复杂度。研究结果验证了所提非侵入式负荷监测方法的有效性及其优越性，对于解决实际低频NILM问题有很大的优势。

Non-intrusive Load Monitoring Method Based on CFSFDP Graph Laplace Algorithm

Non-intrusive load monitoring (NILM) is one of the important development directions of power grid construction in China in the future. In order to overcome the problems of large amount of calculation data and low identification accuracy of traditional NILM methods, a non-intrusive load monitoring method based on CFSFDP (Clustering by fast search and find of density peaks) graph Laplace algorithm was proposed in this paper. Firstly, the power threshold vector and the prior graph structure were constructed using the active power data adopting the CFSFDP algorithm. Then the graph Laplacian quadratic optimal function was constructed by combining the total power signal and graph signal smoothness, and the optimal solution was obtained iteratively by Tikhonov regularization method, so as to realize the reconstruction of graph signal of appliance. Finally, the graph signals were converted into power signals according to the power threshold vector, which enabled non-intrusive load monitoring. The following results were obtained from the simulation analysis of two days of measured electricity consumption data of a real household. 1) The load identification precision of this method for the first day is 100%, and the error between the power consumption proportion of each appliance and the actual power consumption proportion is less than 3%.2) The load identification accuracy of this method for the next day is 90.1%, which is at least 0.8% higher than that of the comparison methods. The decomposition accuracy of a single appliance is more than 91%. Also, the power consumption error is less than 5% and lower than that of the comparison methods.3) When the data sampling interval is increased to 2min, the precision, identification accuracy and decomposition accuracy of single appliance are reduced. But the proposed method gains superior performance and time complexity than the comparison methods. The results verify the effectiveness of the proposed non-intrusive load monitoring method and its superiority for solving practical low-frequency NILM problems.