To address the challenges of low estimation efficiency, inadequate accuracy, and substantial impact of temperature on the state of charge (SOC) estimation for lithium-ion batteries, a novel approach incorporating an improved aquila optimizer (IAO) algorithm with a random forest (RF) algorithm was proposed. In this approach, the IAO algorithm was employed to optimize the hyperparameters of the RF algorithm, thereby enhancing the model's generalization capability and resistance to interference while preventing convergence to local optima. Experimental results reveal that when considering the temperature variations in SOC estimation, the RF model yields a mean absolute error (MAE) of 2.77% and a root mean square error (RMSE) of 3.73%. In comparison, the proposed IAO-RF model achieves a significantly lower MAE of 0.407% and an RMSE of 0.871%, reflecting reductions of 0.85 times and 0.77 times, respectively, compared to the single RF model. Furthermore, the IAO-RF model outperforms other SOC estimation models. These results highlight that even under temperature fluctuations, the IAO-RF model not only enhances estimation precision but also effectively reduces the impact of large errors on overall SOC prediction. The proposed model demonstrates superior accuracy and computational efficiency, providing valuable insights for battery management system (BMS) research.