Accurate prediction of instantaneous power demand during flight is critical for fuel optimization and power coordination control in an aircraft’s energy management system. Traditional power calculation methods rely heavily on solving flight mechanics models and state equations; although they offer high theoretical precision, they struggle to meet real-time requirements and incur significant computational costs due to iterative convergence in complex environments. To overcome these limitations, recent data-driven approaches leveraging artificial intelligence have emerged, aiming to forecast future power demand from historical flight state sequences. However, such time-series models often exhibit inadequate generalization when confronted with complex scenarios or unanticipated changes in flight missions. In this study, we propose a non-sequential modeling framework that treats each flight state as an independent sample, irrespective of temporal order, and applies various machine learning and deep learning algorithms for regression. Comparative experiments involving representative models—namely, Multilayer Perceptron (MLP), Long Short-Term Memory (LSTM) networks, and LightGBM—demonstrate that MLP achieves the most robust performance in mapping high-dimensional flight states to instantaneous power output, offering superior accuracy and generalization. The proposed approach provides a reliable solution for real-time, efficient power prediction in future aircraft energy management systems.