• 主题
高级搜索  >
历史搜索 清空历史
首页> 外文学位 >Lyapunov function based neurocontrollers for a class of deterministic and stochastic problems.
【24h】

Lyapunov function based neurocontrollers for a class of deterministic and stochastic problems.

机译：基于李雅普诺夫函数的神经控制器，用于一类确定性和随机性问题。

获取原文
获取原文并翻译 | 示例
页面导航
  • 摘要
  • 著录项
  • 相似文献
  • 相关主题

摘要

This dissertation focuses on the applications of neural networks in optimal control and adaptive control for both deterministic and stochastic problems.; A two-neural network approach, called the adaptive critic method, is first used as a basic approach to optimally control low-dimensional deterministic problems. However, the optimality of this method is limited to the underlying model of the system. Hence, a Lyapunov based robustness analysis is developed to include uncertainties in the assumed model. It leads to development of another control, called the extra control in this dissertation. This extra control added with the basic control effort from adaptive critic method guarantees good system performance and robust stability under uncertainty. This approach is illustrated through two applications: a nonlinear missile autopilot problem and a miniature helicopter problem.; The second part of this dissertation deals with control of stochastic systems. Dynamic Programming (DP) approach is a more general formulation for optimal control of most systems. However, the overwhelming computational requirements of classical DP algorithms render them inapplicable to most practical problems. A neural network based DP approach is described in this dissertation for optimal stochastic process control. The cost function which is critical in the DP will be approximated by a neural network according to some designed weight-update rule based on temporal difference learning. A Lyapunov based analysis is developed to guarantee an upper error bound between the output of the cost neural network and the true cost. This neural network based DP approach is a combination of adaptive critic method and the extra control design. A motivating scalar example is provided; a more complex retailer inventory management problem is also solved through this approach to show the potential of the developed approach.
机译：本文主要研究神经网络在确定性和随机性问题的最优控制和自适应控制中的应用。首先，将一种称为自适应批评家方法的双神经网络方法用作优化控制低维确定性问题的基本方法。但是，这种方法的最优性仅限于系统的基础模型。因此，开发了基于Lyapunov的鲁棒性分析，以在假设的模型中包括不确定性。它导致了另一个控件的开发，在本文中称为额外控件。这种额外的控制功能加上自适应注释方法的基本控制功能，可确保在不确定情况下具有良好的系统性能和稳定的稳定性。通过两种应用说明了这种方法：非线性导弹自动驾驶问题和小型直升机问题。本文的第二部分讨论了随机系统的控制。动态编程（DP）方法是用于大多数系统最佳控制的更通用的表达方式。但是，经典DP算法的压倒性计算要求使其不适用于大多数实际问题。本文介绍了一种基于神经网络的DP方法，以实现最优的随机过程控制。在神经网络中至关重要的成本函数将由神经网络根据一些基于时间差异学习的设计权重更新规则进行近似。开发了基于Lyapunov的分析，以保证成本神经网络的输出与真实成本之间的误差上限。这种基于神经网络的DP方法是自适应批判方法与额外控制设计的结合。提供了一个激励性的标量示例；通过此方法还解决了更复杂的零售商库存管理问题，以显示已开发方法的潜力。

著录项

  • 作者

    Huang, Zhongwu.;

  • 作者单位

    University of Missouri - Rolla.;

  • 授予单位 University of Missouri - Rolla.;
  • 学科 Engineering Mechanical.
  • 学位 Ph.D.
  • 年度 2002
  • 页码 135 p.
  • 总页数 135
  • 原文格式 PDF
  • 正文语种 eng
  • 中图分类 机械、仪表工业;
  • 关键词

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱：kefu@zhangqiaokeyan.com

京公网安备：11010802029741号 ICP备案号：京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有

  • 客服微信

  • 服务号