基于特征选择与增量学习的非侵入式电动自行车充电辨识方法

为实现从电网侧监测电动自行车违规停放充电行为,减少电动自行车充电火灾事故,在非侵入式负荷识别的基础上,提出一种基于特征选择与增量学习的电动自行车充电辨识方法。首先,根据电动自行车充电实测电流波形,分析负荷特性并列举15种负荷特征。通过半监督Fisher计分与最大信息系数量度特征辨别度与冗余度,采用贪心搜索算法对特征重要性排序并结合排序与辨识结果选择辨识准确性最高的特征子集。然后,基于一类支持向量机增量学习方法,实现电动自行车负荷辨识与分类器在线学习。最后,通过实测数据进行试验,结果表明文中方法可以对电动自行车充电行为准确辨识,验证了算法的有效性。

Non-intrusive Charging Identification Method for Electric Bicycles Based on Feature Selection and Incremental Learning

In order to monitor the illegal parking and charging behavior of electric bicycles from the grid side and reduce fire accidents caused by electric bicycle charging, on the basis of non-intrusive load identification, a charging identification method for electric bicycles is proposed based on feature selection and incremental learning. First, according to the measured current waveform of electric bicycle charging, this paper analyzes the load characteristics and lists 15 load features. The semi-supervised Fisher score and the maximal information coefficient are used to measure the feature discrimination and redundancy, and the greedy search algorithm is used to rank the importance of features. The feature subset with the highest identification accuracy is selected based on the ranking and identification results. Then, based on the incremental learning method for one-class support vector machine, the electric bicycle load identification and online learning of the classifier are realized. Finally, the experiment is carried out with actual measured data, and the results show that the proposed method can accurately identify the charging behavior of electric bicycles, which verifies the effectiveness of the algorithm.