SBIR Phase II: Neural Networks, Youla Parameterization and Reconfigurable Control Systems

9321264 Saeks本项目将开发一种混合频域/神经控制算法，通过使用神经网络技术来调整“优拉设计参数”。这产生了一个保证稳定的控制器，可以实现任何和/或干扰抑制，极点放置，鲁棒跟踪和/或干扰抑制，同步设计，初始值设计等。将神经控制固有的自适应和鲁棒性融入到频域控制器中。强大的（本体）神经网络训练技术可以在控制系统设计问题上发挥作用，允许人们设计一个没有事先限制其顺序的控制器。开发了一种可重构的控制体系结构，其中控制器经过训练以处理多种场景；飞行控制器的备选轨迹、机械臂上的可变负载、电源中的备选电流波形等；在必要时推广到训练集之外的新场景。该方法直接适用于线性多变量系统，并可通过Youla参数化的各种推广推广到非线性情况。*** v是干扰集成检测。广义场景超越可扩展的多样性b C D E F G G I N O P Q R S T U V W X Y Z {} ~ 9321264 Saeks本项目将利用神经网络技术开发一种混合频域/神经控制算法。= H R V b k q {)。​$ (f /。$ .* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *DOT摘要Alicia E. Harris

9321264 Saeks This project will develop a hybrid frequency domain/neurocontrol algorithm by using neural network techniques to adjust the "Youla design parameter." This yields a controller which is guaranteed to be stable and can implement any of and /or distrubance rejection, pole placement, robust tracking and/or disturbance rejection, simultaneous design, initial value design, etc. The adaptivity and robustness intrinsic to neurocontrol is integerated into the frequency domain controller. Powerful (ontogenic) neural network training techniques may be brought to bare on the control system design problem allowing one to design a controller without a prior restrictions on its order. A reconfigurable control architecture is developed where the controller is trained to deal with multiple scenarios; alternative trajectories for a flight controller, varying loads on a robot arm, alternative current waveshapes in a power supply, etc; generaliziang to new scencarios byond the training set where necessary. The approach is directly applicable to linear multivariate systems and extendable, via the vairous generalizations of the Youla parameterization to the nonlinear case. *** v s t disturbanceintegratedetc.generalizingscenariosbeyondextendiblevariousB C D E F G I N O P Q R S T U V W X Y Z \ ^ _ ` a b g h S j k l m n r s t u v w x y z { } ~ 9321264 Saeks This project will develop a hybrid frequency domain/neurocontrol algorithm by using neural net work techniques to . = H R V b k q { ) . $ ( F / . $ . 1 Times Symbol " Helvetica 5 Courier New E O Y ` = H N R 4 V C b U k q { , " h j2 % 9 E ( S R:\WW20USER\ABSTRACT.DOT abstract Alicia E. Harris Alicia E. Harris