Abstract：To address the issue that the loss characteristics of charging device inverters are easily affected by

spatiotemporal changes，leading to a low fit between the identified total loss rate and the actual value，an automated loss identification technique for charging device inverters was proposed that integrates CNN-LSTM（convolutional neural network and long short-term memory）dual encoder feature extraction. CNN was used to extract spatial loss features，while LSTM was employed to extract time-series loss features. By fusing these features，a prediction waveform was generated，and the loss Fourier series and loss magnetic potential were calculated. This enables the determination of current amplitude states under different operating conditions，leading to the construction of an automated identification model. This model identified the magnetic flux distribution state，thereby achieving automated identification of charging device inverter losses. Experimental results show that under additional resistance conditions of 10 Ω，20 Ω，and 30 Ω，the total loss rate of the inverter identified by this technology highly correlates with the actual value，with no significant deviation. The RMSE and MAE are both no more than 0.020.This method effectively improves the accuracy of loss identification in charging device inverters，providing a reliable technical foundation for optimizing device operation strategies，reducing energy loss，and lowering maintenance costs.

        Key words：dual encoder；charging device inverter loss；loss time series；loss waveform；automated

identification

     Format Citation:李涛，宋强，余斌,等. 融合CNN⁃LSTM双编码器的充电设备逆变器损耗自动化识别技术[J].电气传动,2026,56(09):75-80. LI Tao，SONG Qiang，YU Bin,et al.Automated identification technology for charging device inverter losses using a fusion of cnn-lstm dual encoders [J].Electric Drive, 2026,56(09):75-80