Adaptive Identification and Control of Dynamical Systems Using Neural Networks

[9811390]控制领域本质上是跨学科的，从设计、开发和生产一方面延伸到数学另一方面。控制的目的是影响动力系统的行为。在保证稳定性和鲁棒性的同时，在不同的环境条件下实现快速、准确的控制是所有控制系统设计的目标。控制理论中发展最好的部分是关于线性系统的，现代工业中使用的大多数控制器都是基于线性控制原理的。当系统的某些参数未知时，就会出现自适应控制问题。当植物特性已知但明显非线性时，问题的复杂性大大增加，当其某些参数/功能未知或随时间变化时，问题变得真正可怕。目前控制这类系统的方法很少。然而，随着工业应用的日益复杂和技术前沿的不断扩展，这类问题也越来越频繁地遇到。无论是在过程控制和飞机控制等成熟领域，还是在空间技术、机器人技术和制造业等新领域，都是如此。本项目将研究使用神经网络解决此类问题的新方法。本项目由四个部分组成。第一部分讨论了一些与基于神经网络的识别和控制相关的重要问题，这些问题需要进一步研究。第二部分研究了基于模式识别的控制问题，以及神经网络在最优控制中的应用。第三部分重新研究了基于神经网络的控制系统的稳定性问题。PI认为，前三个部分对于更好地理解非线性自适应控制中遇到的困难是必不可少的，并且它们将为第四部分和最后一部分奠定基础，其中包含了工作的主要内容。在这里，详细研究了当时间变化和外部扰动存在时，使用多个模型控制线性和非线性动力系统的问题。

9811390NarendraThe field of control is inherently interdisciplinary in nature and extends from design, development and production on the one hand to mathematics on the other. The objective of control is to influence the behavior of dynamical systems. Achieving fast and accurate control under different environmental conditions, even while assuring stability and robustness, is the aim of all control systems design.The best developed part of control theory deals with linear systems, and most of the controllers used in modern industry are based on linear control principles. When some of the parameters of the system are unknown, we have an adaptive control problem. The complexity of the problem is substantially greater when the plant characteristics are known but distinctly nonlinear, and becomes truly formidable when some of its parameters/functions are unknown or vary with time. Very few methods currently exist for controlling such systems. However, as applications in industry are becoming more complex and the frontiers of technology are being extended, such problems are being encountered with increasing frequency. This is the case both in well established areas such as process control and aircraft control, as well as new areas such as space technology, robotics, and manufacturing. New methods for addressing such problems using neural networks will be studied in this project.This project consists of four parts. The first part deals with some of the important questions related to neural network based identification and control that require further investigation. In the second part, the problem of control based on pattern recognition, and the use of neural networks in optimal control, are studied. The third part re-examines the question of stability of neural network based control systems. The PI believes that the first three parts are essential for a better understanding of the difficulties encountered in nonlinear adaptive control, and that they will set the stage for the fourth and final part, which contains the main thrust of the work. Here, a detailed study will conducted of the problem of controlling both linear and nonlinear dynamical systems using multiple models, when time variations and external perturbations are present.