基于复合特征的非侵入式电力负荷分解与辨识

Non-Intrusive Power Load Disaggregation and Identification Based on Composite Features

摘要: 为减少由于电动自行车违规充电行为导致的火灾事故，文章提出了一种基于数据分离和特征匹配的两阶段非侵入式负荷识别方法。第一阶段，通过复合滑动窗的累积和事件检测算法得到负荷接入点，利用负荷叠加与分离原理从聚合电流中分离出投切负荷的独立负荷信息。第二阶段，将分离出的独立负荷电流与已建立的稳态电流波形库相比较，利用皮尔逊相关系数计算两者的相似度，实现负荷粗辨识；通过特征量化方法提取独立负荷信息的电流幅值特征、功率特征、V-I轨迹特征和谐波特征，与电动自行车的标准特征相匹配，进行负荷精细化识别。实测数据分析结果表明本文所提方法能够有效识别电动自行车的入户充电行为。

Abstract: To reduce fire accidents caused by the illegal charging behaviors of electric bicycles, this paper proposes a two-stage non-intrusive load monitoring (NILM) method based on data separation and feature matching. In the first stage, a cumulative sum (CUSUM) event detection algorithm with a composite sliding window is adopted to obtain load connection points, and the principle of load superposition and separation is utilized to separate the independent load information of switched loads from the aggregated current. In the second stage, the separated independent load current is compared with the established steady-state current waveform library, and the Pearson correlation coefficient is used to calculate the similarity between the two to achieve coarse load recognition. A feature quantization method is employed to extract the current amplitude features, power features, V-I trajectory features and harmonic features of the independent load information, which are then matched with the standard features of electric bicycles for refined load recognition. The analysis results of field measurement data demonstrate that the method proposed in this paper can effectively identify the in-home charging behaviors of electric bicycles.