面向未知非线性关联系统主动容错控制的自适应动态规划方法

The unknown dynamic model of the nonlinear interconnected system is identified by combining the input/output data of the isolated subsystem with recurrent neural network, meanwhile, the reference states of interconnected subsystems are employed to estimate the interconnection term. The reasonable performance index function, which is estimated via critic neural network, is established by considering the interconnection term, the identification error, the replacement errors of reference states as well as disturbances, then the decentralized optimal control scheme is proposed by an identifier-critic structure. The faulty subsystem is further decoupled into an actuator fault subsystem and a sensor fault subsystem. Taking the robustness and the fault sensitivity into account, the fault threshold generator is designed to judge subsystems in fault or not. Hereafter, the fault observer is designed by using dual linear matrix inequality (LMI) technique to provide a less conservation sufficient condition to stabilize the observation system. Constructing an improved performance index function which reflects all fault types, and the fault tolerant control problem is transformed into an iterative optimization problem via adaptive dynamic programming through the system transformation, and then the adaptive dynamic programming scheme of active fault tolerant control is presented for unknown nonlinear interconnected systems. The effectiveness of the proposed schemes is demonstrated by combining the numerical simulation with experiments of hardware in loop/hardware. This project is not only expected to extend the application range of adaptive dynamic programming theory, but also provide a new method of fault tolerant control, as well as the great theoretical and application value.

采用子系统输入/输出数据结合递归神经网络辨识未知非线性关联系统模型，并利用关联子系统的参考状态结合神经网络估计非匹配未知关联项。建立考虑关联项、辨识误差、参考状态取代误差、干扰等因素的性能指标函数，并采用评判神经网络对其进行估计，研究基于辨识-评判结构的分散最优控制方法。将故障子系统分解为执行器故障子系统与传感器故障子系统，兼顾系统鲁棒性与故障敏感性设计鲁棒自适应阈值发生器；基于双线性矩阵不等式技术设计故障观测器，给出保守性更小的使观测系统稳定的充分条件。构造反映各种故障模态的改进性能指标函数，通过系统变换将容错控制问题转化为基于自适应动态规划的迭代调节问题，研究面向未知非线性关联系统主动容错控制的自适应动态规划方法。采用数值仿真结合半实物/实物实验的方式验证所提各方法的有效性。本项目研究不仅可拓展自适应动态规划理论的应用范围，还可为容错控制提供新的解决方案，具有重要的理论和工程应用价值。

本项目面向未知非线性关联系统，基于自适应动态规划理论、最优控制理论与容错控制理论，针对分散最优控制、故障诊断、容错控制及其验证等问题开展研究。1) 采用子系统输入/输出数据，并利用关联子系统的期望信息取代子系统关联项中相应的实际信息，结合神经网络实现对未知关联系统的在线辨识，并分析在线辨识方法的合理性。2) 针对无故障非线性关联系统，研究了基于策略迭代算法的分散最优镇定控制、基于改进性能指标函数的受扰系统镇定控制、基于数据与确定性策略梯度的无模型系统控制、基于辨识-评判结构的分散跟踪控制以及时变时滞关联系统的局部近优控制等方法。3) 采用自适应估计、故障观测器、有效因子估计与神经网络估计等方法对故障进行在线检测与辨识，并将相应方法用于容错控制器的设计。4) 针对非线性互联故障系统，提出了基于改进性能指标函数的容错控制、基于故障在线补偿的容错控制、基于有效因子融合的分散容错控制、基于前馈神经网络的饱和故障系统容错控制等方法。5) 采用数值仿真算例、可重构机械臂系统、双倒立摆系统、办公建筑用电系统等为例，验证所提各种方法的有效性。本项目的研究不仅拓展了自适应动态规划理论的应用范围，还为容错控制提供了新的解决方案，具有重要的理论和工程应用价值。