航空发动机LQR控制的模糊神经网络方法

Aeroengine LQR Control Based on Fuzzy-Neural Networks

针对线性二次型调节器LQR在航空发动机多变量控制中存在的存储量需求太大的问题,提出了相应的自适应神经网络模糊控制方法。根据某型发动机飞行包线内给定工作点的线性化模型,分别设计控制器,并将分别设计的控制器用自适应神经网络模糊推理的方法进行综合,使之成为一个非线性的控制器,由此可以得出其它工作点的LQR设计结果。该方法能够在一定程度上弥补LQR控制的缺陷,仿真实例表明了其有效性。

To solve the problem of Linear Quadratic Regulator (LQR) in aeroengine multivariable control,a modified LQR method based on a kind of fuzzy-neural networks was presented.First,we chose some representative operation points in the engine flight envelope,and designed the LQR controller for each operating point separately.Then,Adaptive-Network-based Fuzzy Inference System (ANFIS) was utilized to synthesize each linear controller to make a nonlinear controller.The inputs of ANFIS are flight altitude and Mach number,and the output is a feedback matrix,which represents the design results.The design of other operation points in the engine flight envelope can be conveniently deduced by the inference system.The ANFIS was trained offline,so we concluded that the method could compensate the short comings of the LQR control.Simulation results were given for a specific turbofan.The design process was analyzed,and the results prove the validity of the method.