Fuzzy-Neural Sliding Mode Control of Uncertain Systems: A Lyapunov Theory Approach

提出的研究目标是将神经网络、模糊系统和遗传算法结合到不确定或非线性动态系统的滑模控制器、滑模状态估计器和滑模辨识器的设计中。将提供一个模糊建模的范例，允许系统地构建模糊模型，用于控制器和状态估计器的设计。对控制器和状态估计器的稳定性和保证性能进行了分析，并在地面车辆仿真模型上进行了验证。利用遗传算法对控制器和估计器的参数进行优化。在实际中，控制器和被控对象都受到各种非线性约束，如增益的硬边界、有限能量或有限切换速度，这些都必须在实际的控制器设计中加以考虑。此外，由于对参数值的不了解或建模过程中的不准确性，设计者必须应对工厂模型中的不确定性。本文采用确定性方法对不确定动态系统进行控制、辨识和状态估计。将研究连续时间滑模神经辨识器的自适应算法，并开发新的变结构滑模模糊控制器和状态估计器。然后，将所提出的结构集成到自组织模糊神经滑模跟踪控制器中。神经网络和模糊逻辑控制器在闭环应用中取得了相当大的成功。然而，这些应用虽然非常成功，但对于控制变量振幅有限的不确定系统没有保证稳定性的证据。在本文的研究中，将Lyapunov的直接方法、Hahn的Lyapunov方法的扩展和LaSalle的不变性原理用于模糊神经滑模控制和识别结构的稳定性和保证性能分析。提出的控制器、估计器和标识符将在最近开发的地面车辆模型上进行测试，其中包括横向重量转移和轮胎模型。该模型适用于车辆动态性能的实时评价。所提出的研究成果将有助于基础控制理论和智能车辆控制系统的发展。

The objective of the proposed research is to incorporate neural networks, fuzzy systems and genetic algorithms into the design of sliding mode controllers, sliding mode state estimators and sliding mode identifiers of uncertain or nonlinear dynamical systems. A paradigm for fuzzy modeling will be provided that allows for systematic construction of fuzzy models for the purpose of the controllers and state estimators' design. The controllers and state estimators' stability and their guaranteed performance will be analyzed and then tested on a simulation model of a ground vehicle. Optimization of the controllers and estimators' parameters will be achieved using genetic algorithms. In practice, the controller as well as the plant are subject to various nonlinear constraints like hard bounds on gains, limited energy, or finite switching speeds that must be taken into account in a realistic controller design. In addition, due to lack of knowledge of parameter values or inaccuracies in the modeling process, the designer must cope with uncertainties in the plant model. In this project, a deterministic approach to the control, identification and state estimation of uncertain dynamical systems is taken. Adaptation algorithms for continuous-time sliding mode neural identifiers will be studied and novel variable structure sliding mode fuzzy controllers and state estimators will be developed. Then, the proposed structures will be integrated into self-organizing fuzzy-neural sliding mode tracking controllers. Neural network and fuzzy logic controllers have been used with considerable success in closed-loop applications. However, these applications, though very successful, have no proofs of guaranteed stability for uncertain systems with control variables limited in amplitude. In the proposed research, the direct method of Lyapunov, Hahn's extensions of the Lyapunov method and LaSalle's Invariance Principle will be used in the stability and guaranteed performance analyses of fuzzy-neural sliding mode control and identification structures. The proposed controllers, estimators and identifiers will be tested on the recently developed ground vehicle model that includes lateral weight transfer and tires' models. This model is suitable for the evaluation of the vehicle dynamical behavior in real-time. The results of the proposed research will contribute to the basic control theory as well as to the intelligent vehicle control systems.