A hybrid onboard model for aero-engine direct thrust predictive control

Abstract Direct thrust control can markedly enhance thrust regulation accuracy and unlock the full performance potential of aero-engines. To improve both real-time capability and precision, we propose a hybrid adaptive onboard predictive modeling framework, termed DNN-PSM-SVM. In this approach, a deep neural network captures strong nonlinearities to refine accuracy, while steady- and dynamic-deviation models based on PSM and SVM reduce computational complexity. A Kalman filter further enhances adaptability, avoiding heavy nonlinear calculations and significantly improving real-time performance. Leveraging this model within a predictive control scheme, unmeasurable parameters such as thrust and surge margin are estimated in real-time, enabling accurate thrust control even under component degradation. Simulation results show that the method outperforms conventional predictive control, achieving steady-state accuracy below 0.06 % and improving real-time performance by nearly an order of magnitude. Unlike sensor-based control, it maintains precise thrust regulation despite engine degradation.