【作者】 翟军勇；

【导师】 费树岷；

【作者基本信息】 东南大学 ， 控制理论与控制工程， 2006， 博士

【摘要】 在实际控制问题中,由于被控对象在不同工况下系统的参数或结构的变化,传统的自适应控制算法是无能为力。针对传统自适应控制和现有多模型自适应切换控制理论和方法中存在的问题,研究面向复杂系统的多模型切换控制新理论和新方法。本文对多模型建模与控制的若干问题作了研究和探索。主要研究内容包括:针对被控系统在不同工况下的模型参数突变,系统暂态响应特性较差,提出基于在线学习的多模型自适应控制方法。应用动态模型库技术来建立多模型,并证明该算法能够保证闭环系统的稳定性和跟踪误差的渐近收敛性。仿真结果表明所提出的建模方法和相应的多模型自适应控制器使系统的动态响应品质得到了明显的改善。基于神经网络强大的学习能力和非线性逼近能力,给出面向复杂系统的神经网络多模型切换控制方法。采用最近邻聚类学习算法对样本分类,并利用RBF神经网络进行离线建模。系统运行时在线检测系统当前状态,若超出现有各子模型所构成的状态空间,利用在线神经网络学习新状态并建立新的子模型加入动态模型库中,从而增强系统的鲁棒性。仿真结果表明该算法的有效性。针对传统多模型自适应控制中子模型数量过多的问题,给出一种基于递阶结构的多模型自适应控制算法。将整个控制系统分为基本工况级和控制模型级的两层递阶结构。在系统运行过程中,由常规自适应模型和重新赋值自适应模型在线自动地建立多模型及相应的控制器。该方法有效地减少了子模型数量和计算时间。将最小方差控制技术、神经网络的逼近能力及多模型切换控制技术相结合,给出一种基于RBF神经网络动态补偿的多模型控制方法。利用李亚普诺夫函数方法推导出网络权值的自适应调整律。采用具有积分性质的切换指标函数作为切换法则和最小方差的控制方法构成多模型自适应控制器。该算法有效地消除不确定引起的控制误差。更多还原

【Abstract】 Due to the parameters or the structure changes of the system to be controlled under the different operating mode, the traditional adaptive control algorithm is helpless in many practical control problems. New theories and new methods are studied for complex systems multiple models switching control in this paper, which aims at the problem of conventional adaptive control and multiple models adaptive control in existence. Here, the dissertation mainly discusses issues on multiple models based modeling and control of complex nonlinear systems. The main contributions of this dissertation are summed as follows:A novel adaptive switching control algorithm based on multiple models is proposed to improve the dynamical response performance of plants with large parameters variations under different operating modes. At the same time dynamic model bank is applied to establish models bank without the prior system information. The closed-loop system stability and track error convergence asymptotically are proved. The simulation results have confirmed the efficacy of the proposed methods.A multiple models adaptive switching control based on neural network is proposed, which aims at the problem of conventional adaptive control and multiple models adaptive control in existence. We adopt the closest cluster algorithm to classify the samples, and then utilize RBF neural network’s strong learning and nonlinear approximates abilities to model off-line. Meanwhile dynamic model bank technology is applied to establish multi-model. Once the real time system’s states go beyond the space, which is formed of all existing sub-model, then a new state is learnt through online neural network and set up a new model that is added in the dynamic model bank in order to improve the dynamic system’s transient response and robustness. Finally, simulations are given to demonstrate the validity of the proposed method.Multiple models adaptive control based on hierarchical structure is presented, which aims at the problem of many sub-models in conventional multiple models adaptive control. The system to be controlled is divided into basic operating-condition level and control model level. Multiple models and corresponding controllers are automatically established on-line by the conventional adaptive model and a reinitialized one. The proposed method can reduce the number of更多还原