Accurate prediction of RUL (Remaining Useful Life) of lithium-ion batteries is important to ensure stable and safe operation of electric vehicles. However, limitations in extracting local features and obtaining long-term serial relationships existed when using neural network-based methods. In view of these challenges, a joint DPCNN (Deep Pyramid Convolutional Neural Networks)-Transformer network is presented to improve the accuracy for RUL prediction of Li-ion batteries. DPCNN is good at capturing local features, and the pyramid Pooling layer can extract features both locally and globally. Transformer performs well when dealing with long sequences. Through Transformer’s self-attention mechanism, the model can better capture long-term dependencies can be well captured by the model. The presented method allows it to take both local and global information into account when learning, capturing a variety of features in the data more comprehensively, thus improving prediction performance. Through the experiments on NASA dataset and CACLE dataset, our method has lower relative error (RE) and root mean square error (RMSE) can be achieved, and also the mean absolute error (MAE) in the remaining charge/discharge counts when compared to other neural network methods like RNN, CNN and LSTM, which proves the effectiveness and superiority of the proposed method.