To rapidly and efficiently evaluate the state of health (SOH) of substation backup power batteries and address the issue of low estimation accuracy in data-driven methods due to the lack of actual operational data, this paper proposes a SOH estimation method for substation batteries that combines electrochemical characteristics and Gaussian process regression (GPR). Traditional studies that use characteristic parameters obtained from single aging experiments struggle to accurately reflect the actual aging conditions of lead-acid batteries used in substation backup power. Starting from the electrochemical essence of the battery, this method designs float charging and cyclic aging experiments to collect electrochemical impedance spectroscopy (EIS) data under different aging mechanisms. Subsequently, highly representative electrochemical characteristic parameters are extracted using Pearson correlation analysis and grey relational analysis. The combination of these two experimental aging characteristics more closely approximates the actual aging characteristics of the battery, effectively improving the quality and efficiency of the training data and reducing the amount of training data required. Finally, these extracted characteristic parameters are used to train the GPR model to achieve accurate SOH estimation for actual substation batteries. The results show that the absolute error (AE) in estimating the SOH of randomly selected substation batteries is less than 0.08, with an average absolute error (MAE) of 0.033 0 and a root mean square error (RMSE) of 0.038 6. This method does not require the collection of actual data and can effectively estimate the SOH of substation batteries with a small amount of experimental data.