A multivariable adaptive control method for aeroengine with H∞ performance considering engine output limitation protection based on fully adjustable Neural Network

Abstract

Aeroengines, being highly nonlinear systems subject to external disturbances, present a challenge for conventional control methods in achieving satisfactory performance. The radial basis function neural networks possess the capability to effectively model highly approximate nonlinear functions, making them suitable for implementation in aeroengine control. However, the radial basis function (RBF) neural network is a local approximation model that inadequately incorporates historical information. Additionally, the tracking error of neural networks can also impact control effectiveness. To address this issue, this paper proposes a multivariable control method for aeroengines with H_∞ performance based on fully adjustable Long Short-term Memory-Radial Basis Function Neural Network (LSTM-RBFNN). Firstly, a factual mathematical model of an aeroengine is established based on the principles of aerodynamics and thermodynamics. Secondly, the paper proposed an LSTM-RBF neural network architecture to enhance prediction performance and proposed using a finite time perturbation estimator to gauge the approximate error of neural networks, thereby acquiring immeasurable approximative error information. Then, in order to mitigate the influence of neural network tracking error on the loop, the estimated errors are incorporated into the primary loop controller, and a static state disturbance full feedback H_∞ controller is proposed. Then, leveraging the LSTM-RBF architecture proposed in this study, we further employ Taylor expansion of nodes to derive the adaptive law of the neural network and propose a novel neural network adaptive law that incorporates tracking error estimated by the disturbance observer into the adjustment mechanism to enhance tracking performance. The stability of this adaptive law has been proven based on Lyapunov function analysis. And considering the output limitation protection problem of aeroengine, a virtual command value based limitation protection method is proposed for the above control method, and it is proved that this limitation method does not damage the stability and design process of the original control method. Finally, numerical simulation and hardware in loop simulation experiments were conducted, and the results showed that under the same adaptive gain of the neural network, the performance indicators (IAE, IATE, σ) could be reduced by more than 50% compared to other methods, which demonstrates superior tracking and disturbance rejection performance achieved by our proposed method.