Performance seeking control of adaptive cycle engine based on Bayesian optimization

According to the performance seeking control of adaptive cycle engine in multiple operating modes, an engine performance optimization control strategy based on surrogate model and Bayesian optimization was proposed to reduce the number of engine model calls and calculation time, and avoid local optimization problems. This method built an adaptive cycle engine surrogate model based on the high-precision component model and the Gaussian process regression. It used penalty functions and indicator functions to transform the optimization problem with constraints into an unconstrained problem. The performance seeking control system based on the adaptive cycle engine model was verified with three optimization modes: minimum fuel flow at constant thrust, minimum turbine temperature at constant thrust, and maximum thrust at maximum dry and full afterburner throttle settings. The simulation results showed that the Gaussian surrogate model constructed with a small number of samples significantly reduced the number of engine model calls, effectively cut down the amount of calculation, and avoided the problem of falling into a local loop during the engine model calling process; the Bayesin optimization algorithm used an active learning strategy to independently increase sample points and update the model based on the convergence condition evaluation agent model; the Bayesian optimization algorithm with global search characteristics can overcome engine performance problems. The optimization algorithm overcame the disadvantage of relying on manual experience, providing an effective solution for engine performance optimization. Optimization results showed 1501.27 N enhancement in maximum thrust mode, 0.38% reduction in minimum fuel consumption mode and 7.9 K reduction in minimum turbine temperature mode for adaptive cycle engine with a core-driven fan respectively.