【作者】 赵俊；

【导师】 陈建军；

【作者基本信息】 西安电子科技大学 ， 机械制造及其自动化， 2009， 博士

【摘要】 现实事物中绝大多数对象都是包含噪声干扰的非线性系统,基于经典控制理论、现代控制理论的传统控制方法往往是针对线性系统设计的,对包含噪声干扰的强非线性、复杂时变系统的应用具有较大的限制。随着人工智能技术的不断发展,以模糊系统、神经网络为代表的智能产物显示出对复杂非线性系统强大的处理能力,一系列基于智能控制理论及方法的控制系统被不断提出和改进,在对复杂对象的控制问题上取得了重大的突破和丰硕的成果。然而,由于各种智能产物的基础理论发展仍不成熟,在应用各种智能控制方法时存在许多值得改进的地方。本论文拟针对基于智能方法的先进控制系统设计提出若干新的参考方法和改进应用方法,具体工作包括以下内容:1.对智能控制理论的背景及发展状况进行了综述,评述了智能控制领域的主要研究方法和获得的成果,阐述了基于智能方法的控制系统综合设计研究的意义和工程实用价值。2.提出一类以模糊神经网络和PID神经网络组成的模糊神经PID网络;提出一种基于混沌优化机制的粒子群优化算法,设计了混沌优化与粒子群结合的两步方案。将上述方法用于控制系统设计,具体构成为:模糊神经PID网络用作控制器,优化策略为带混沌机制的粒子群算法离线优化和误差反传算法在线调整相结合的方法;被控对象为确定性典型非线性和惯性时滞对象。3.提出一种基于最小二乘支持向量机建模的自适应智能PID控制系统。控制系统具体构成:控制器及其优化算法采用模糊神经PID网络和改进粒子群算法的方案;引入最小二乘支持向量机用于离线建模,将控制系统拓展到能处理具有未知特性的不确定对象的控制问题。4.提出一种基于改进蚁群算法优化的大时滞对象神经网络控制系统。控制系统具体构成:控制器采用模糊神经PID网络,其离线优化采用一种改进的蚁群优化算法,在线时采用误差反传算法调整;利用最小二乘支持向量机辨识来获取系统下一离散时间步的预估值,对不确定大时滞对象进行离线辨识和在线辨识来处理时滞和不确定性问题。对空调房间对象进行了控制仿真。同时,设计了基于径向基函数神经网络的空调系统模型参考自适应控制系统,给出了前向型神经网络控制系统设计的一般性方案。5.针对航空发动机对象,提出一种综合模糊推理、神经网络自适应和PID控制各自优点的控制系统。控制系统具体构成:模糊神经PID网络用作控制器,其参数优化策略采用改进蚁群算法离线优化和误差反传在线调整的方法;最小二乘支持向量机用于系统的离线和在线辨识,其参数优化选取采用交叉验证的方法。对某型航空发动机设计点处的线性和非线性模型进行了控制仿真。6.针对航空发动机加速过程的控制问题,通过结合多种智能方法,提出了一种基于分类转换策略的控制系统。控制系统具体构成:模糊神经PID网络用作控制器,提出一种改进的量子粒子群算法离线优化其参数;利用标称模型将加速过程中发动机大范围不确定模型划分为若干小偏差不确定模型,作为未知控制对象;离线时利用最小二乘支持向量机结合主成分分析方法对小偏差模型进行分类和辨识训练,在线时根据系统实时数据利用分类器自动选择相应的小偏差模型,同时利用误差反传算法实时调整控制器参数跟踪期望信号;分类器和辨识器参数分别采用交叉验证和量子粒子群算法优化选取。基于模式识别思想和智能神经网络控制实现了一种新颖的依据系统信息实时选择对象模型的非线性PID控制策略。对某型航空发动机的加速过程进行了控制仿真。7.研究一类系统参数在很大范围内变化的不确定对象,为克服传统鲁棒控制方法的保守设计缺陷并进一步改善系统的性能,提出一种分类转换控制策略:在已知系统参数变化上下界前提下,基于类似分段线性化的思想,将系统进行分割;对于分割后的多个小偏差范围模型,利用最小二乘支持向量机结合主成分分析方法进行分类;对每类模型分别设计滑模控制器,并利用一种改进量子粒子群优化算法离线优化构造近似最佳切换函数,同时利用径向基函数神经网络结合误差反传算法在线调整切换项增益的方法降低系统的抖振;在线时根据系统实时数据利用分类器自动选择相应的小偏差模型和滑模控制器,完成相应的控制作用;为提高最小二乘支持向量机的分类及泛化性能,利用改进量子粒子群算法优化其惩罚因子和核参数。基于以上策略和优化配置,对控制系统进行了设计与仿真。8.针对以模糊神经自适应方法为核心的未知非线性系统控制问题,以常规静态模糊神经网络控制结构为基础,分别就控制器、辨识器、优化算法三个方面展开研究。以模糊神经PID网络作为控制器,最小二乘支持向量机为辨识器,利用改进量子粒子群算法离线优化控制器参数和改进粒子群算法优化辨识器的相关参数,最后通过对系统的稳定性分析将控制系统逐步完善,完成对基于模糊神经网络方法的自适应控制系统中各个环节的改进。对某热交换非线性对象进行了控制仿真。最后对论文的主要工作进行了概括性的总结,阐述了所获得的一般性结论。列出了论文工作的主要创新之处,对后续的研究工作进行了展望。更多还原

【Abstract】 Nonlinear characteristics and noise disturbance exist in most of the actual systems. The conventional control methods based on classical control theory and modern control theory are mainly designed for linear systems, which can hardly be applied to strongly nonlinear and complex systems. As the artificial intelligence technologies continue to develop, products of these technologies represented by fuzzy system and artificial neural network show the powerful processing capability to complex nonlinear systems. A series of advanced control systems based on intelligent control theories and methods are continuously put forward and improved, great breakthrough and plentiful fruits are derived for the control problem of complex systems. However, due to the immaturity of the foundation theory of intelligent control,there are still many aspects and key points need to be improved when applying the control methods. In view of the combined design of the advanced intelligent control system in future, the main research work of this dissertation can be described as follows:1、This chapter gives an overview of the background and the development of intelligent control theory, sums up the main research methods and achievements of intelligent control field, illustrates the significance and practical value in the study of the combined design of intelligent control system.2、A fuzzy neural PID controller consists of a fuzzy neural network and a PID neural network is proposed. A partical swarm optimization (PSO) algorithm based on chaos optimization is designed. The control system is designed based on the work above, the system scheme includes: the fuzzy neural PID network is used as the controller; the parameters of the controller are optimized by the mixed learning methods integrating offline particle swarm optimization algorithm combined with chaos strategies of global searching ability and online BP algorithm of local searching ability; a typical nonlinear object and a time delay object are used as controlled plant.3、A self-adaptive intelligent PID control system is presented based on least squared support vector machine (LS_SVM). LS_SVM is used as the identifier for uncertain objects. The proposed control system scheme includes: the controller and the optimization algorithm configuration are the same as the last chapter; by introducing LS_SVM, the control system is extended to an improved scheme which can handle the objects with uncertainty.4、A neural network control system for time-delay objects is presented based on an improved ant colony algorithm (ACO). The proposed control system scheme includes: the controller uses the fuzzy neural PID controller; the parameters of the controller are optimized by the mixed learning methods integrating offline chaotic ant colony optimization (CACO) and online error back propagation algorithm; LS_SVM is used as identifier, and it is trained offine and online to obtain the forecasting value of the next step time of the discrete system. By applying the proposed control system, the simulation is done with a air conditioning room object. Also, the model reference adaptive control system for the whole air conditioning system is designed based on radial basis function (RBF) neural network. The general designing method of the control system using feed-forward neural network is described in details.5、In view of complicated, undetermined and strongly nonlinear aeroengine object, a novel control scheme integrating the merits of fuzzy inference, neural network adaptation and simple proportional-integral-derivative (PID) method is presented. The proposed control system scheme includes: the controller uses the fuzzy neural PID controller; the parameters of the controller are optimized by the mixed learning methods integrating offline chaotic ant colony optimization (CACO) and online error back propagation algorithm; LS_SVM is used as identifier, and it is trained offine and online; the parameters of LS_SVM is optimized by cross-validation method; the simulation is done with a aeroengine object at the designed work condition.6、A novel control scheme consists of several intelligent control strategies for aeroengine acceleration process is proposed. The proposed control system scheme includes: the controller is constructed by a fuzzy neural PID network; the parameters of the controller are optimized by offline quantum-behaved particle swarm optimization (QPSO) with chaos strategy combined by online error back propagation tuning; in process of acceleration, the model with the wide range of uncertainty is partitioned by designed nominal model and several models with the small range of uncertainty are derived; these models are classified and identified by LS_SVM offline. In online situation, the designed model with the small range of uncertainty is selected automatically by LS_SVM based on system data, and the controller is tuned by error back propagation simultaneously; the parameters of the classifier are optimized by cross-validation and the identifier by QPSO. The whole system with the model selected by online system data is designed by nonlinear PID strategy based on pattern recognition and intelligent neural network; the simulation is done with an aeroengine object in acceleration process.7、In order to overcome the conservation of conventional robust control methods, a new classifying and switching strategy based on LS_SVM for the control of uncertain system with the parameters varying in a wide range is proposed. The proposed control strategy includes: the original system model is divided into several models with small range of uncertainty; these models are classified by LS_SVM combined with principal component analysis (PCA) offline; for each model, the sliding-mode controller (SMC) with its gain tuned by RBF neural network is designed to reduce the chattering phenomenon effectively, and the QPSO algorithm with chaos strategy is designed and applied to adjusting the parameters of the controller so as to construct an optimized switching function. In online situation, each model with the designed SMC is selected automatically by LS_SVM based on system data; the parameters of LS_SVM are optimized by QPSO with chaos strategy to improve the performance of classification and generalization. In final, the system scheme is designed by the proposed method.8、Considering the fuzzy neural adaptive control method for undetermined nonlinear system, the control scheme are studied and improved based on the conventional fuzzy neural network control scheme. The controller, identifier and optimization algorithm of the scheme are designed respectively by the improved new methods. The proposed control system scheme includes: the fuzzy neural PID network is used as the controller, and LS_SVM as the identifier; the controller is optimized by offline QPSO with chaos strategy combined by online error back propagation tuning; the parameters of LS_SVM are optimized by PSO with chaos optimization; the stability of the improved scheme is discussed to finish the whole design in conclusion; the simulation is done with a heat exchanger object.In final, the main research work is summarized. As a whole, the conclusions derived from this dissertation is illustrated. The innovation points and the future outlook of the research work is given.更多还原