Aiming at the problem that the capacity regeneration phenomenon affects the state of health ( SOH) prediction accuracy of lithium-ion batteries, an energy-weighted Gaussian process regression (EWGPR) of empirical mode decomposition (EMD) method is proposed. This method regards the capacity recovery phenomenon as the energy projection of the capacity decay process of lithium-ion battery. The energy distribution is obtained by EMD decomposition and the sample weights are calculated according to the energy distributions. Then the SOH prediction model of lithium-ion battery based on EWGPR is established. The experimental simulation results on the NASA lithium-ion battery datasets show that the EWGPR algorithm has higher accuracy and adaptability than the basic GPR algorithm, and the root mean square error (RMES) for single-step and multi-step predictions are decreased by more than 3% and 10%, respectively.