An Unsupervised Non-intrusive Electric Bicycle Charging Detection Method

In recent years, more fires caused by unsafe electric bicycle charging have occurred in China. In this paper, we propose a novel method for detecting the unsafe charging of electric bicycles, based on unsupervised non-intrusive load monitoring technology. First, the characteristics of the typical electric bicycle battery charging power curves are introduced. Then, we generate a new time-series signal leveraging both active power and reactive power, and perform pre-process using bilateral filtering and state transition removal. Finally, the piecewise linear representation is used to segment the time-series signal, and the segments are identified by pre-set rules as charging loads. The proposed method is validated on real smart meter power readings for three scenarios, showing promising overall detection results.