Unsupervised identification and status assessment for electric bicycle charging load

Electric bicycle (EB) is a common short-distance transportation means in China, Recently, fire incidences caused by thermal runaway of EB batteries occur frequently nationwide, drawing people's attention. Although policies have been made that EB charging is forbidden in residential buildings, people may not obey due to various reasons. Therefore, identification of EB charging load (EBCL) in residential buildings, especially the abnormal batteries with fire danger, is beneficial to public safety. To meet this urgent need, an unsupervised EBCL identification and battery status assessment method based on non-intrusive load monitoring technology is proposed in this paper. At first, the specifications of typical EB batteries are introduced with the demonstration of EBCL signals for batteries in either normal or abnormal status. Then, pre-processing steps including signal transformation, multiple filtering steps, and state transition removal, are proposed. Next, the signal sub-sequences related to EBCL characteristics are obtained via piecewise linear representation. Finally, post-processing steps are proposed to refine the EBCL identification results and detect abnormal batteries. Validation is carried out on the power readings containing charging loads for various EB battery types, collected from real Chinese households. The experimental results show the proposed method outperforms two state-of-the-art benchmarking NILM methods in various metrics.