During the actual operation of electric vehicles, the dynamic changes in the health status of power batteries are directly related to driving safety and the reliability of driving range. The superposition of complex road conditions, fluctuations in ambient temperature, and random charging and discharging behaviors makes it difficult for traditional monitoring methods to accurately capture the subtle characteristics of battery aging. By leveraging the nonlinear fitting capabilities and deep feature extraction abilities of computer deep learning, correlations between health status and multidimensional parameters can be established from vast amounts of operational data. Through optimizing model architecture and selecting input features, prediction adaptability under various operating conditions can be improved, reducing deviations caused by insufficient data or noise. This provides technical support for the intelligent upgrade of battery management systems and offers references for the rational planning of battery cascade utilization to a certain extent.