Lithium-ion batteries are the core energy source for New Energy Vehicles and Energy Storage Systems, making the prediction of performance degradation and lifespan a critical BMS research priority. Accurate State-of-Health (SOH) quantification is vital for system safety and energy efficiency. To overcome the challenges of strong nonlinearity and time-series dependency in battery cycling, we propose an innovative LSTM-GPR multivariate hybrid prediction model.The model utilizes the Long Short-Term Memory network (LSTM) to automatically extract deep degradation features from multi-dimensional time-series data (voltage, current, temperature). Gaussian Process Regression (GPR) is subsequently employed for residual correction and quantifying prediction confidence.Tested on the NASA battery dataset, the model exhibits excellent SOH prediction performance: minimum R2is 0.9975, and maximum RMSE and MAE are only 0.00499 Ah and 0.00463 Ah, respectively, significantly outperforming CNN and SVR models. This LSTM-GPR hybrid model, combining high accuracy with quantifiable reliability, offers a robust online SOH prediction tool for BMS, providing a new theoretical basis for battery life management.