Accurate estimation of lithium battery capacity is the basis for its safe and reliable operation. However, the lithium battery capacity attenuation mechanism is not clear, and its capacity shows nonlinear attenuation during use, making it extremely difficult to accurately estimate the lithium battery capacity. In order to solve these problems, this paper proposes a method to estimate the capacity of lithium batteries by using constant current charging data and a one-dimensional convolutional neural network model. Firstly, two aging feature sequences are extracted from the constant current charging stage： the constant current charging capacity and the time interval of equal charging voltage rise. Secondly, in order to make full use of the measured charging information and extract the key capacity attenuation factors in aging feature sequences, a new one-dimensional convolutional neural network capacity estimation model is constructed. Finally, the proposed capacity estimation model is verified and analyzed on the aging data set of lithium batteries in Maryland. The experimental results show that the proposed method can achieve an accurate and robust estimation of battery capacity on the complete constant current charging curve and partial constant current charging curve, which is better than other capacity estimation methods. The root mean square error （RMSE） and mean absolute error （MAE） can be controlled within 0.68% and 0.50%. Furthermore, after transfer learning with a small amount of date, the proposed model can accurately estimate battery discharge capacity at other discharge rates. After transfer learning for the first 50 cycles, the root mean square error of the model at a 0.5C discharge rate is only 0.58%.

lithium batteries  /  convolutional neural networks  /  transfer learning  /  capacity estimation  /  constant current charging  /  deep learning