Direct learning control for an Aero-engine based on Adaboost-LSSVM

As a typical complex thermodynamic systems with strongly nonlinear and rapid time varying characteristics in the whole flight envelope, the design of a flexible control strategy for aero-engines has always been a challenging problem. This paper proposes a direct learning control method of the aero-engine base of the adaboost-LSSVM, which can directly to design the LPV model reference controller of IO data onto the aero-engine without parameterizing, identifying, and completing the statespace implementation of the system's LPV model. After representing the controller for the IO form, the entire design process is transformed into a convex optimization problem, which is further solved by the LSSVM. After that, an adaptive enhancement algorithm, which integrates a strong model consisting of multiple basic models, is employed. Not only improves the control accuracy, it also effectively reduces the degree of overfitting. In addition, due to LSSVM, the proposed method also achieves a satisfactory trade-off between the control accuracy and computational efforts. Finally, simulation results of a turbofan engine control system verify the effectiveness of the proposed method.