In order to improve the estimation accuracy of lithium battery state of health (SOH), a novel estimation method combining the optimal reconstruction of Health Indicators and RIME-optimized support vector regression (RIME-SVR) is proposed. First, three measurable Health Indicators are extracted from the charging and discharging process of lithium batteries, and their correlation with SOH is verified using the Pearson method. Subsequently, the complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) algorithm is employed to decompose and reconstruct the health indicators. The optimal reconstruction method is determined through experimental validation, effectively reducing the interference of data noise and capacity recovery on SOH estimation. Finally, an SVR estimation model optimized by the RIME algorithm is established. The experiments are conducted using NASA battery degradation data. The results show that compared with particle swarm optimization (PSO) and artificial bee colony (ABC) algorithms, RIME exhibits faster convergence speed and stronger global search capability when optimizing SVR parameters, significantly enhancing model performance. Moreover, the lithium battery SOH estimation model based on the optimal reconstruction of health indicators and RIME-SVR outperforms other models in the comparative experiments in terms of three indicators, achieving higher estimation accuracy and fitting degree. When the optimally reconstructed health indicator Dtv1+Ti1+Tdv1 is used as input, the model’s average mean absolute error (MAE) is below 0.37, root mean squared error (RMSE) is below 0.55, and the coefficient of determination is higher than 0.92, indicating good universality and robustness.