LIBs exhibit considerable potential in energy storage applications; however, prolonged usage can result in performance degradation, manifesting as reduced maximum capacity, diminished power output, and elevated internal impedance. The SOH serves as a quantitative metric indicative of a battery's overall aging status, correlating with capacity deterioration. Because of the complexity of measuring individual parameters and the complexities of internal battery electrochemistry, precise assessment and prediction of SOH remain difficult. This study presents a comparative analysis of hybrid models for predicting SOH utilizing time series data. The BiLSTM model is effectively augmented through the integration of SOH battery data characteristics in conjunction with Attention Mechanism (AM) and Positional Encoding Attention (PEA). Experimental findings demonstrate that the BiLSTM-PEA model exhibits superior performance. The implementation of PEA significantly enhances the processing of time series SOH battery data, resulting in improved computational efficiency. PEA could assist the model in concentrating on the general trend in data of SOH batteries. © 2024 IEEE.