Adaptive Dynamic Programming for Continuous-Time Systems and Networked Agents on Graphs

提案编号：eccs -0801330提案标题：图上连续时间系统和网络代理的自适应动态规划名称：Lewis， Frank L.PI机构：德克萨斯大学阿灵顿分校目标和方法本研究的目的是通过使用计算智能领域的技术设计更好的高性能自动反馈控制结构来提高人类工程系统的性能。该方法是在连续时间系统中使用自适应动态规划（ADP）方法，这是一种尚未完全开发或用于控制系统设计的技术。π吗?s先前的工作表明，连续时间系统的ADP为高性能控制系统设计提供了一个新的框架，从中出现了反映人类大脑自然多时间尺度结构的控制功能的多时间尺度分离。知识价值。生物结构中的调节系统是实时的最优和自适应的，允许在线学习达到最佳性能。然而，人工控制系统要么是离线优化设计，要么是实时自适应方法。本研究将利用ADP技术开发统一的自动控制设计方法，该方法既具有最优性，又具有实时适应性。将提供性能的数学上严格的证明。更广泛的影响。这项工作将开发出更好的反馈控制结构，从而为飞机、发动机、工业过程等提供更好的性能。它将有助于将计算智能和控制系统社区结合在一起的理论基础。将通过在机电实验室设备和三区电网试验台上的实施，对基于adp的技术和标准自适应鲁棒控制技术进行全面比较。通信图系统的合作ADP算法的发展将允许更有效地控制电力系统网络和车辆编队。正在进行的研究课程开发活动将有助于将研究成果转化为UTA的教育和培训。这笔拨款将允许扩展现有的自动化和机器人研究所（ARRI）项目，支持工程领域的女性、美国本科生研究和K-12外展。屡获殊荣的SBIR计划将促进技术转移到美国工业。

Proposal Number: ECCS-0801330Proposal Title: Adaptive Dynamic Programming for Continuous-Time Systems and Networked Agents on GraphsPI Name: Lewis, Frank L.PI Institution: University of Texas at ArlingtonObjective and ApproachThe objective of this research is to improve the performance of human engineered systems by designing better high performance automatic feedback control structures using techniques from the Computational Intelligence field. The approach is to use adaptive dynamic programming (ADP) methods for systems working in continuous-time, a technique not yet fully developed or exploited for control systems design. The PI?s prior work shows that ADP for continuous-time systems provides a novel framework for high-performance control system design from which emerges a multi-timescale separation of control functionality mirroring natural multi-timescale structures in the human brain.Intellectual Merit.Regulation systems in biological structures are both optimal and adaptive in real-time, allowing on-line learning to achieve the best performance. However, manmade control systems are either optimal off-line designs or real-time adaptive methods. This research will use ADP to develop unified automatic control design methods that are both optimal and adaptive in real-time. Mathematically rigorous proofs of performance will be provided. Broader Impacts.This work will develop better feedback control structures that will facilitate better performance for aircraft, engines, industrial processes, etc. It will contribute to theoretical foundations bringing together the Computational Intelligence and Control Systems communities. Comprehensive comparisons between ADP-based techniques and standard adaptive & robust control techniques will be made through implementations on electro-mechanical laboratory equipment and on a three-area power network testbed.Development of cooperative ADP algorithms for systems on communication graphs will allow more effective control of networks of electric power systems and vehicle formations.Ongoing activities in research curriculum development will facilitate transferring research results to education and training at UTA. This grant will allow extension of existing Automation and Robotics Research Institute (ARRI) programs supporting women in engineering, US undergraduate student research, and K-12 outreach. The award-winning SBIR program will facilitate technology transfer to US industry.