The state of health (SOH) evaluation and remaining useful life (RUL) prediction for lithium-ion batteries (LIBs) are crucial for health management. This paper proposes a novel sequence-to-sequence (Seq2Seq) prediction method for LIB capacity degradation based on the gated recurrent unit (GRU) neural network with the attention mechanism. An improved particle swarm optimization (IPSO) algorithm is developed for automatic hyperparameter search of the Seq2Seq model, which speeds up parameter convergence and avoids getting stuck in local optima. Before model training, the complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) algorithm decomposes the capacity degradation sequences. And the intrinsic mode function (IMF) components with the highest correlation are employed to reconstruct the sequences, reducing the influence of noise in the original data. A real-cycle-life data set under fixed operating conditions is employed to validate the superiority and effectiveness of the method. The comparison results demonstrate that the proposed model outperforms traditional GRU and RNN models. The predicted mean absolute percent error (MAPE) in SOH evaluation and RUL prediction can be as low as 0.76% and 0.24%, respectively.

Lithium titanium oxide (LTO) batteries offer superior performance compared to graphite-based anodes in terms of rapid charge/discharge capability and chemical stability, making them promising candidates for fast-charging and power-assist vehicle applications. However, commonly used battery models often struggle to accurately describe the current–voltage characteristics of LTO batteries, particularly before the charge/discharge cutoff conditions. In this work, a novel electrical model based on the solid-phase diffusion equation is proposed to capture the unique electrochemical phenomena arising from the diffusion mismatch between the positive and negative electrodes in high-power LTO batteries. The robustness of the proposed model is evaluated under various loading conditions, including constant current and dynamic current tests, and the results are compared against experimental data. The experimental results for LTO batteries exhibit remarkable alignment with the model estimation, demonstrating a maximum voltage error below 3%.