<p>To achieve better accuracy in SOH and RUL estimation, this paper proposes a hybrid neural network framework based on Nonlinear AutoRegressive with eXogenous inputs (NARX) and bidirectional long short-term memory (BiLSTM). Specifically, firstly, by analyzing the battery characteristic curve, three indirect health indicators (HIs) are extracted: constant voltage charging current, constant current charging voltage, and constant current charging time. In order to reduce the size and noise of the extracted HIs, a stacked autoencoder method is proposed to reduce the size and noise of the extracted HIs, and the correlation between the HIs and capacity is analyzed by using the correlation method. Secondly, a hybrid neural network is proposed to establish the SOH and RUL estimation framework. The NARX model embeds BiLSTM memory, which considers the context information of the sequence in the time expansion model, providing a shorter path for the propagation of gradient information, reducing the long-term dependence on recurrent neural networks. Finally, the proposed model is validated on different datasets, and the experimental results showed that the SOH estimation error is limited to &lt; 2% MAE, and the RUL estimation error is controlled within ± 3 cycles, which has good advantages and estimation ability.</p>