内容简介：Adaptive Dynamic Programming (ADP) has received increasing attention recently. ADP scheme is a design that approximates dynamic programming in the general case, i.e., approximates optimal control over time in noisy, nonlinear environments. There are many engineering problems in practice which can be formulated as cost maximization or minimization problems. Dynamic programming is a very useful tool in solving these problems. However, it is often computationally untenable to run dynamic programming due to the backward numerical process required for its solutions. Over the years, progress has been made to provide approximate solutions to dynamic programming. The idea is to approximate dynamic programming solutions by using neural networks to approximate the cost function. The methodology is a very useful tool for building intelligent agents/controllers in almost any environment.

    This talk will review the theorectical development of ADP. Details about the training of the neural networks used in the present design will also be presented. The pole balancing (inverted pendulum) problem will be used as the benchmark in this presentation to show the applicability of ADP.

个人简介：刘德荣，生于1963年，吉林省白城人。于1982年从华东工学院(现南京理工大学)毕业并获机械工程学士学位，于1987年从中国科学院自动化研究所毕业并获自动控制理论及应用硕士学位，于1994年从美国圣母大学(University of Notre Dame)毕业并获电机工程博士学位。从1982年至1984年，在中国北方工业公司国营向阳仪表厂工作(吉林省)。从1987年至1990年，在中国科学院研究生院无线电电子学部任教(北京市)。从1993年至1995年，在美国通用汽车公司研究开发中心工作(密西根州-Warren, Michigan)。从1995年至1999年，在斯蒂文斯理工学院电机与计算机工程系任助教授(新泽西州-Hoboken, New Jersey)。从1999年开始，在依利诺斯大学(芝加哥)电机与计算机工程系工作。先后任该校助教授、终身职副教授，现任该校计算智能实验室主任、电机与计算机工程系研究生部主任、以及电机与计算机工程系和计算机科学系两个系的终身职正教授。2005年，他因在非线性动态系统和递归神经网络方面作出的贡献而被选为电气与电子工程师学会会士(IEEE Fellow)。他当时是美国仅有的几位获得此项殊荣的副教授之一。