During the charging process of electric vehicles, the conversion between AC and DC power, as well as the transformation between electrical and chemical energy, inevitably leads to electrical energy loss. To accurately calculate the magnitude of this loss with limited measurement equipment, this paper proposes a charging pile efficiency calculation theory based on Kmeans-ADE-LSTM. This method adopts a memory neural network to train the model with historical charging data, deriving a model relationship between the DC side electrical data, temperature data and charging efficiency, and then validates it with new charging data. Innovatively, the collected charging data is subjected to Kmeans clustering analysis before efficiency calculation, and the processed data is then used for subsequent neural network training. Additionally, to address the difficult problem of selecting the mutation coefficient in traditional differential evolution algorithms, this paper employs an efficient global optimization algorithm-the differential evolution algorithm-to adaptively process the mutation coefficient and optimally select the hyper parameters of the LSTM. Based on the correlation between charging efficiency and time, etc., the charging conversion efficiency is calculated in conjunction with the LSTM neural network. The comparative experimental results demonstrate that the proposed Kmeans-ADE-LSTM charging pile charging efficiency prediction method has a high level of predictive accuracy.