The state of charge (SOC) prediction of aviation nickel-cadmium batteries is a critical technology for ensuring the safe operation of aircraft. To address the issues of insufficient accuracy and poor environmental adaptability in traditional prediction models, this study proposes a hybrid WOA-RF prediction model that combines the whale optimization algorithm (WOA) with random forest (RF). Firstly, an initial prediction model is constructed based on the random forest regression algorithm, leveraging its multi-decision tree ensemble advantage to handle nonlinear features. Secondly, the whale optimization algorithm is introduced to globally optimize the core hyperparameters of the random forest, resolving the inefficiency of manual parameter tuning and thereby enhancing the model’s prediction accuracy and generalization capability. To validate the model’s performance, discharge cycle experiments were conducted under different temperature conditions (20 ℃, 0 ℃, -10 ℃, -20 ℃), and the prediction results of the WOA-RF model were compared with those of traditional RF, backpropagation neural network (BPNN), support vector regression (SVR), as well as particle swarm optimization RF (PSO-RF) and genetic algorithm-optimized RF (GA-RF) models. The experimental results show that under standard temperature conditions, the WOA-RF model achieves a mean absolute error (MAE) of 1.22%, a coefficient of determination (R2) of 0.986, and a root mean square error (RMSE) of 1.56%, outperforming the comparison models. In low-temperature environments, the WOA-RF model maintains an MAE below 1.5%, an RMSE below 1.8%, and an R2 above 0.975, demonstrating stronger environmental robustness. The conclusion indicates that the WOA-RF model effectively improves the accuracy and stability of SOC prediction, making it particularly suitable for monitoring the state of nickel-cadmium batteries under extreme aviation operating conditions. This provides reliable technical support for battery management systems.