•  
  •  
 

Keywords

electric bicycle, deep learning, non-intrusive load monitoring, feature extraction, attention mechanism

Abstract

The growing popularity of electric bicycles is accompanied by the risk of charging-related fires，calling for prompt effective monitoring.Non-intrusive load monitoring （NILM ） provides a key technical approach for this purpose.A deep learning identification algorithm based on label-guided attention is proposed for non-intrusive identification of electric bicycle charging loads.Specifically，a feature sample library is constructed by collecting high-frequency power data from electric bicycles and various household appliances.Then，deep features are extracted using a one-dimensional convolutional neural network （1DCNN ），and feature representation is enhanced by combining feature and position attention mechanisms.Furthermore，sequential dependencies are captured using a bidirectional recurrent neural network （BRNN ） to achieve accurate load forecasting.Finally，simulation analysis is conducted using the publicly available plug-load appliance identification dataset （PLAID ） and actual data of electric bicycle charging loads to verify the effectiveness of the proposed method.The results show that the proposed method performs excellently in the identification task，delivering results superior to those of the existing models.

DOI

10.19781/j.issn.1673-9140.2026.04.011

First Page

125

Last Page

135

Download

Share

COinS