【摘要】 为杜绝安全隐患，利用V-I轨迹和改进MobileNetv2模型对入户充电行为进行在线辨识。设计实验场景，从采样率选取、迁移学习、泛化性和不同网络对比4个方面验证模型性能，最后把模型部署到上位机和K210芯片上。上位机系统在电动自行车单独充电时准确识别，当充电行为和常用家庭负载混合运行时，识别准确率达到98%以上。更多还原

【Abstract】 To prevent the safety hazards, the V-I trajectory features and improved MobileNetv2 model are used for the online identification of the household charging behavior.The experimental scenarios are designed to validate the model performance from four aspects: sampling rate selection, transfer learning, generalization, and comparison of different networks.Finally, the model is deployed to the computer and the K210 chip.The online recognition system based on the upper computer can accurately identify electric bicycles when charging separatly, and the recognition accuracy is over 98% when charging behavior is mixed with commonly used household loads.更多还原