Abstract: The data-driven-based capacity prediction is essential for lithium-ion battery health management to extend its lifetime. However, most of the state-of-the-art methods are based on laboratory data analysis, which can not reflect the aging characteristics of the actual vehicle battery under complex conditions. Therefore, a hybrid model based on support vector machine and improved Gaussian process is designed to accurately predict the capacity of electric vehicle battery by using real vehicle data. Firstly, the capacity data is extracted from the real-time operation data set of vehicles by using the sliding window ampere-hour integration method, and the ensemble empirical mode decomposition method is designed to divide the battery capacity into two parts: long-term degradation trend and short-term fluctuation. Then, the support vector machine and the improved Gaussian process are designed respectively to model the two components, and the results are fused to obtain the final capacity prediction value. Extensive experiment results on three vehicles validate the effectiveness of the proposed hybrid prediction model, with higher accuracy demonstrated than the commonly used methods.