HNN-based generalized predictive control for turbofan engine direct performance optimization

• Published in: Aerospace science and technology Vol. 112; p. 106602
• Main Authors: Zhou, Xin, Huang, Jinquan, Lu, Feng, Liu, Jie, Wang, Chen
• Format: Journal Article
• Language: English
• Published: Elsevier Masson SAS 01.05.2021
• Subjects: Constraint nonlinear optimization; Direct performance optimization control; Generalized predictive model; Hopfield neural network; Turbofan engine; Direct performance optimization control; Hopfield neural network; Turbofan engine; Generalized predictive model; Constraint nonlinear optimization
• ISSN: 1270-9638; 1626-3219
• DOI: 10.1016/j.ast.2021.106602

The aero-engine performance optimization control is to find the appropriate control variables to maximize or minimize the comprehensive index of one or more performance parameters while ensuring its safe operation. Traditional performance seeking control is mainly based on indirect control, with complex structure and large error. This paper develops an HNN (Hopfield Neural Network)-based generalized predictive control strategy for turbofan engine direct performance optimization. With a Hopfield network as the optimization algorithm and a CARIMA model based on generalized prediction criterion as the optimization model, an optimal controller is designed. Together with the identification module and the self-adaptive module, an integrated performance optimization control system is formed. Namely, the adaptive model calculates the unmeasurable performance parameters of the engine for the CARIMA model identification. Then HNN performs the iterative operation to obtain the optimal control variables for a given objective. The main contribution of this paper is to propose a direct performance control structure to avoid the nonlinear conversion of control reference; meanwhile, the linear CARIMA model is combined with HNN algorithm to build a neural network for solving constrained nonlinear programming problems, and mutation operation is introduced to improve its global search capability, which is evident in the minimum turbine temperature optimization mode. The method is evaluated on a turbofan model from the aspects of model verification, performance optimization control simulation at multiple operating points and under degeneration conditions. The results confirm the satisfactory performance of the proposed strategy at the selected points and conditions. And compared with the traditional algorithms, the computer calculation time is greatly shortened.