Accurate and efficient lithium-ion battery capacity prediction plays an important role in improving performance and ensuring safe operation. In this study, a novel lithium-ion battery capacity prediction model combining successive variational mode decomposition (SVMD) and aquila optimized deep extreme learning machine (AO-DELM) is proposed. Firstly, SVMD is used to divide capacity signal and it improves short-term trend prediction, especially for capacity growth that occurs during the degradation process. Secondly, the DELM network outperforms other networks in efficiently extracting time-dependent features, and it is more accurate than other standard ELM-based methods. The AO algorithm is used to optimize the parameters of the DELM training process for the problem of sensitivity to initial weights. Finally, experiments are conducted to validate the predictive performance of the proposed model based on NASA and CALCE lithium-ion batteries discharge capacity decay sequences. The MAE (0.0066Ah, 0.0044Ah), RMSE (0.0113Ah, 0.0078Ah), MAPE (0.44%, 0.82%) are effectively reduced and the R2 (98.94%, 99.87%) are better than the prediction performance of other hybrid models.