This paper presents a lithium-ion battery state-of-health (SOH) estimation method based on a long short-term memory neural network. The proposed algorithm uses multi-input for the current state of the battery, which are voltage, current, temperature and state-of-charge (SOC). In addition, it reflects the historical state of the battery such as difference of current and temperature by varying the input sequence length, for a more accurate SOH estimation. To verify the proposed algorithm, the accuracy value when using the error function is checked. Then, the algorithm selects the appropriate sequence length and analyzes the system. The system improves estimation accuracy through the feedback process. Furthermore, the proposed method is compared to other learning methods to support the validation. The results highlight the selection of the appropriate sequence length can improve the accuracy of the estimation and show that the battery SOH estimation with the optimal sequence length is great effect on the performance through RMSE 0.1425, MAE 0.1084, and MAPE 0.8658%.