The intrinsic properties of lithium-ion batteries can lead to a substantial reduction in their charge-discharge cycles when exposed to prolonged unstable conditions, such as overcharging, over-discharging, and overheating. In response to the challenges associated with accurately predicting battery lifespan and prolonging battery service life, this study introduces a novel methodology for evaluating the state of health (SOH) of lithium-ion batteries. This methodology combines the robust feature extraction capabilities of the Transformer model with the advanced temporal modeling proficiency of the Long Short-Term Memory (LSTM) neural network. Additionally, it utilizes Bayesian Optimization (BO) algorithms to automatically determine the optimal hyperparameter settings, thereby significantly improving the accuracy and efficiency of SOH assessments for lithium-ion batteries. The experimental validation of the proposed approach is conducted using a publicly available lithium-ion battery dataset from the National Aeronautics and Space Administration (NASA). The findings indicate that the BO-Transformer-LSTM model outperforms traditional assessment methods, achieving a significant enhancement in assessment accuracy. By addressing the limitations of conventional assessment techniques, this approach not only improves the accuracy and efficiency of SOH evaluations but also offers substantial theoretical implications and practical applications.