基于反演设计的块控型极值搜索系统一体化控制方法研究  

Research on Integrated Control Design for Block Control Extremum Seeking System Based on Backstepping Design

作　　者：左斌[1] 李静[2] 胡云安[1] 

机构地区：[1]海军航空工程学院控制工程系,烟台264001 [2]海军航空工程学院战略导弹工程系,烟台264001

出　　处：《自动化学报》2011年第9期1114-1129,共16页Acta Automatica Sinica

基　　金：国家自然科学基金(60674090)资助~~

摘　　要：针对极值搜索控制系统(Extremum seeking control systems,ESCSs)设计中,极值搜索算法与控制器采取单独设计时易导致系统难以发挥其最佳性能,而现有的一体化设计方法却存在需要根据被控对象和具体的极值搜索算法进行不同形式的一体化建模的问题,以块控型的极值搜索控制系统为研究对象,提出了一套通用的极值搜索控制系统的一体化控制方法.首先针对块控型极值搜索控制系统,采用反馈线性化设计思想,构建出系统的伪虚拟控制量;然后以极值搜索算法得到的搜索变量作为其输入量,设计多层神经网络(Multilayer neural networks,MNNs)逼近由近似模型与实际模型之间的差异而导致的误差项、状态变量的极值和极值的变化率,同时运用自适应参数和鲁棒项函数抵消神经网络逼近误差的影响;最后利用反演控制方法求取出系统的虚拟控制量和实际控制量.此一体化控制方法确保系统的状态跟踪误差、输出量与其极值之间的误差、极值搜索变量的跟踪误差以及神经网络各参数的估计误差均有界且指数收敛至系统原点的一个有限邻域内,且理论分析和仿真结果都验证了此方法的有效性.In the process of designing the traditional extremum seeking control systems （ESCSs）, the independent design method may lead the designed ESCSs to be difficult to get their optimal performance. Whereas, the current integrated design of ESCSs is an intricate method that needs to accomplish the integrated modeling for different controlled plants and the concrete extremum seeking algorithms （ESAs）. To solve those problems, a universal integrated control method for block control ESCSs is proposed. Firstly, based on feedback linearization technology, the pseudo virtual control law is designed for block control ESCSs. Secondly, with the search variables of ESAs as the inputs, multilayer neural networks （MNNs） are used to approximately cancel the error between an approximate model and real model, and approximate the extrema of the states and the differential of the extrema. At the same time, an adaptive parameter and continuous robust terms are adopted to minify the influence of the MNNs construction error. Lastly, the virtual control laws and the final control law are derived using backstepping control method. This integrated control scheme guarantees that all of signals, which include the tracking errors of the states, the errors between the outputs and their extrema, the tracking errors of the search variables, and the estimation errors of MNNs＇ parameters, are the ultimate boundedness of the closed-loop system. Theoretical analysis and simulation results show the validity of this integrated control method.

关 键 词：极值搜索算法 一体化设计 反演控制 神经网络 反馈线性化 

分 类 号：TP13[自动化与计算机技术—控制理论与控制工程]