【作者】 袁鸯；

【导师】 孙健国；

【作者基本信息】 南京航空航天大学 ， 航空宇航推进理论与工程， 2005， 硕士

【摘要】 本文从数学模型和控制器两方面着手研究提高航空发动机控制系统自适应性的方法。包括机载自适应实时模型研究和神经网络控制研究两部分。 本文建立发动机机载自适应实时模型的思想认为:发动机的任何非额定工况都必然导致其输出参数变化,输出参数会偏离它们的额定值产生输出偏离量。通过设计卡尔曼滤波器和神经网络映射模块的方法可以在线实时地获得这些输出偏离量,并将其用来修正按额定特性建立的发动机部件级模型的输出,最后再利用修正后的模型参数,经过一系列非线性计算即可获得发动机非额定工况下的性能量(推力、喘振裕度等)的准确值,从而具备对发动机非额定工况的自适应能力。 关于神经网络控制,本文主要研究了基于 BP 网络整定的 PID 控制和神经网络并行控制两种方案。前者利用神经网络具有的非线性逼近能力,通过对系统性能指标的学习来实现 PID 控制参数的最佳组合;后者将神经网络作为前馈控制器与 PID 反馈控制器共同起控制作用。两种方案都充分利用了神经网络在线学习调整权值的能力,增强了控制器对对象特性变化的自适应能力。本文最后还对 CMAC 网络模型进行了学习和研究,并尝试着将其应用于航空发动机线性模型的控制,取得了较好的控制效果,为进一步研究做准备。更多还原

【Abstract】 To improve the adaptive capability of aeroengine control system, the research on mathematic model and controller has been conducted in present thesis. The work includes two parts: research on adaptive modeling and research on neural network control. The main idea of setting up adaptive real-time model of aeroengine believes that outputs of aeroengine will bias their nominal values in any case of off-nominal work. These biases include the off-nominal information of engine. We can design Kalman filter module and neural network module to obtain these biases on-line and real-time. Then these biases can be used to modify the outputs of onboard component-level model which is set up with nominal characteristic. After modification, outputs of onboard model are the same as those of the real engine, and the real time onboard model has the ability of adaptation. In the second part of thesis, two neural network control schemes are discussed: the PID control based on BP neural network identification, and the parallel control based on BP neural network and PID. The first scheme uses the nonlinear mapping ability of neural network to realizes the best parameters combination of PID. The second scheme uses BP neural network as forward controller and PID as feedback controller to realize parallel control. Both schemes take full advantage of tuning neural network’s weights on-line to enhance the adaptive capability of controller. The controller using CMAC neural network is also studied in present thesis. It’s used to control linear models of certain type aeroengine. The simulation results of the control system using CMAC neural network are satisfying.更多还原