A Non-intrusive Method Based on Deep Learning for Abnormal Electricity Consumption Detection of Electric Bicycles

Abnormal electricity consumption of electric bicycles has given rise to many severe accidents (e.g., explosion and fire accidents). Primary causes of these accidents are users’ incorrect charging behavior and lack of stipulated safety standard designed for different charging devices. From utility’s perspective, it is of great importance to detect abnormal electricity consumption of electric bicycles in a non-intrusive way considering the customers’ privacy concern. Therefore, this paper proposed a non-intrusive method based on deep learning for abnormal electricity consumption detection of electric bicycles. Firstly, charging curve and charging process of electric bicycles are studied. Then customers’ electricity consumption data is analyzed, the missing values are filled in and the outliers are removed to prepare dataset. Afterwards, convolutional neural network (CNN) model is constructed and trained to identify the abnormal data. Finally, results of CNN model are compared with deep neural network (DNN) and other machine learning techniques in order to demonstrate the effectiveness of this method.