随着科技水平和生产需求的不断提升，实际系统的规模日益庞大，结构也变得更加复杂。互联非线性系统是一类典型的大规模系统，针对其包含的各个子系统分别设计控制律，构成一组控制器以实现整个互联系统闭环稳定的分散控制方式已被广泛采用。随着对系统控制品质需求的不断提升，现有控制方法已无法满足互联系统在控制精度、响应速度以及优化性能等方面更高的要求。此外，系统固有组件自身特性的限制以及安全运行的需求使得系统在运行过程中受到状态约束，这也给控制系统的设计带来了困难。同时，基于时间触发的控制方法使用固定采样周期来更新控制信号，导致了频繁的通信和计算，造成了计算资源和通信资源的浪费。因此，在考虑系统状态约束的情况下，对于广泛存在的大规模互联非线性系统，研究一种资源节约的分散镇定方法，具有重要的理论意义和实际应用价值。
    本文基于自适应动态规划（Adaptive dynamic programming，ADP）理论，分别针对含有常值型和不等式型两种状态约束的互联非线性系统研究其镇定问题，主要内容如下：
    （1）针对一类含有常值型全状态约束的互联非线性系统，提出了一种基于ADP 的分散镇定方法。引入边界函数对原系统进行坐标变换，将状态约束系统转化为无约束系统。将转化后的互联系统分解为一系列相互关联的独立子系统，进而构建子系统的代价函数，通过求解各个独立子系统的最优调节问题实现原互联系统的分散镇定。构建局部评判神经网络并采用策略迭代算法近似求解哈密顿–雅可比–贝尔曼(Hamilton-Jacobi-Bellman, HJB) 方程，得到近似最优镇定律，并通过Lyapunov 稳定性理论证明了所提镇定方法可使闭环互联系统和局部评判神经网络估计误差动态最终一致有界。通过仿真案例，验证了所提分散镇定方法的有效性。该方法引入的边界函数可以根据状态约束直接对系统进行变换，无须对系统进行扩维处理，一定程度上避免了变换过程使系统模型变得更加复杂。
    （2）针对一类含有不等式型状态约束的互联非线性系统，提出了一种基于ADP 的事件触发分散镇定方法。引入松弛函数对状态约束方程进行变换得到松弛变量，进而对原系统进行增广。针对增广子系统构建代价函数，并设计局部评判网络对其进行估计，基于梯度下降法设计了具有稳定补偿的局部评判网络权重学习律，进而得到分散镇定律。根据Lyapunov稳定性理论设计触发机制，并证明了所提出事件触发分散镇定方法可使闭环互联非线性系统稳定。通过对比仿真验证了所提方法可确保互联非线性系统的闭环稳定性的同时，满足状态约束条件。该方法能够处理不等式型的状态约束，且不需要初始可容许镇定律，引入的事件触发机制节约了通信和计算资源。
    最后，对本文的研究工作进行了总结，并对未来的研究工作进行了展望。

 As technology advances and production demands increase, the scale of practical systems is growing and becoming more complex. Nonlinear interconnected  systems are a typical example of large-scale systems. A widely adopted approach to address them is to design controllers for each subsystem separately, forming a group of controllers to achieve decentralized control and stabilize the entire  closed-loop interconnected system. However, as the demand for control quality increases, existing control methods are no longer sufficient to meet higher requirements in terms of control precision, response speed, and optimization performance. Moreover, the inherent limitations of system components and the requirement for safe operation impose constraints on the system during operation, adding difficulty to control system design. Additionally, time-triggered control methods, which use fixed sampling periods to update control signals, which leads to frequent communication and computation and results in wastage of computational and communication resources. Therefore, considering the constraints on system states, researching a resource-efficient decentralized stabilization method for widely present large-scale interconnected nonlinear systems holds significant theoretical and practical value.
    This article based on Adaptive Dynamic Programming (ADP) theory, investigates the stabilization problems of nonlinear interconnected systems with two types of state constraints: constant constraints and inequality constraints, the main content is as follows:
    (1) For a class of interconnected nonlinear systems with constant state constraints, a decentralized stabilization method based on ADP is proposed. By introducing barrier functions for coordinate transformation, the system's state-constrained nature is transformed into an unconstrained form. The interconnected system is decomposed into a series of interrelated independent subsystems, and the cost function of each subsystem is constructed. This transforms the stabilization problem of the interconnected system into the optimal adjustment problem of each independent subsystem. Local critic neural networks are constructed, and a policy iteration algorithm is used to approximate solutions to the Hamilton-Jacobi-Bellman (HJB) equation, obtaining an approximate optimal stabilization law. The proposed method is shown, through Lyapunov stability theory, to ensure that the closed-loop interconnected system and the estimation error of the local critic neural network are uniformly ultimately bounded. Simulation examples verify the effectiveness of the proposed decentralized stabilization method. The introduced boundary functions can directly transform the system based on state constraints without the need for system dimension expansion, thus avoiding making the system model more complex.
    (2) For a class of interconnected nonlinear systems with inequality state constraints, an event-triggered decentralized stabilization method based on ADP is proposed. By introducing relaxation functions to transform the state constraint equations into relaxed variables, the original system is augmented. A cost function is constructed for each augmented subsystem, and local critic networks are designed to estimate them. Based on gradient descent method, a stable compensation learning law for the weights of local critic networks is designed, thereby obtaining a decentralized stabilization law. A new triggering mechanism is designed based on Lyapunov stability theory, and it is proved that the proposed event-triggered decentralized stabilization method ensures the stability of the closed-loop interconnected nonlinear system. Comparative simulation examples demonstrate that the proposed method ensures the stability of interconnected nonlinear systems while satisfying state constraints. This method can handle inequality state constraints without requiring knowledge of initial admissible control laws, and the introduced event-triggering mechanism also saves communication and computational resources.
    Finally, the research work of this paper is summarized, and future research directions are outlined.