Accurate state-of-health (SOH) estimation is essential for the reliable health management of lithium-ion batteries. However, in practical applications, health indicators (HIs) extracted from battery cycling data may form incomplete degradation sequences because valid operating segments satisfying consistent HI definitions are not always available, resulting in nonuniform observation intervals and a mismatch between discrete sequence steps and the actual degradation process. To address this issue, this study proposes an SOH estimation method based on a prereconstructed dynamic feature accumulation closed-form continuous-time network (PRDFA-CfC). The prereconstruction module enhances multi-scale temporal dependencies within the HI sequence, while dynamic feature accumulation preserves historical degradation information. The actual intervals between consecutive health observations are explicitly incorporated into the closed-form continuous-time state update to construct a time-aware latent degradation representation for SOH estimation. Experiments are conducted on a publicly available lithium-ion battery ageing dataset, in which incomplete degradation sequences are constructed by randomly removing health observations from the original cycling sequence. Comparative results show that the proposed PRDFA-CfC achieves an RMSE of 0.4274% and an MAE of 0.3616%, outperforming LSTM, GRU, Transformer, ODE-RNN, and DFA-CfC. The results demonstrate the effectiveness of the proposed method for SOH estimation under incomplete degradation sequences.

Lithium-ion battery,state of health,incomplete degradation sequence,continuous-time neural network