Abstract: Lithium-ion batteries are widely used in power storage, but their capacity and power degradation problems highlight the importance of health state(SOH) assessment and degradation prediction. Although machine learning has been applied to battery performance simulation, state estimation and other fields, most studies lack the quantification of prediction uncertainty. Battery degradation involves the coupling of multiple factors such as SEI film growth and electrode cracking. Health management needs to solve six core problems such as SOH estimation and remaining life(RUL) prediction. Although deep learning has the advantage of self-extraction of features, it faces the defects of small sample over-fitting and black box model. Probabilistic deep learning can alleviate data conflict problems through confidence quantification. In practical applications, the process needs to be followed: Construct input and output models and data sets, select probabilistic algorithms that take into account both prediction accuracy and uncertainty quantification, and output prediction indicators after training and verification. In the future, it is necessary to focus on the development of interpretable probabilistic models that integrate physical mechanisms to improve the reliability of battery health prediction and promote the development of intelligent management technology for energy storage systems.