This study investigates the application of five machine learning and deep learning methods—CatBoost, LightGBM, XGBoost, MLP, and DNN—for real-time fuel consumption prediction in the turboelectric hybrid propulsion system. The complexity of the system, characterized by high internal coupling, nonlinear dynamics, and multivariable input-output relationships, poses significant challenges for traditional mathematical modeling approaches. Using a dataset of 15,000 real instances of steady-state engine data, we evaluated the performance of each method in terms of Mean Squared Error (MSE) and Root Mean Squared Error (RMSE). Our findings indicate that LightGBM achieves the highest accuracy, while the deep learning methods, particularly DNN, exhibit lower performance due to increased model complexity. In contrast, LightGBM, which has a simpler model, has the best performance, with only a 2.7% error in its performance on the test set. The results underscore the importance of balancing model complexity with prediction accuracy and inference speed. Therefore, in this paper, a lightweight fuel consumption estimation model based on machine learning with high real-time performance and high accuracy is implemented. This study provides a valuable reference for the selection and optimization of predictive models in designing energy management strategy for the turboelectric propulsion system.