【作者】 龚固丰；

【导师】 章兢；

【作者基本信息】 湖南大学 ， 控制科学与工程， 2009， 博士

【摘要】 论文基于免疫系统原理来研究非线性系统控制领域两大重要课题:非线性系统模型辨识及其预测控制。免疫系统是一个具有强大学习能力的分布式动态鲁棒系统,在免疫系统中种类有限的抗体能够识别种类繁杂并且处于不断进化中的抗原。抗体识别抗原的这种机制显示免疫系统具有强大的自学习、自组织能力及良好的自适应性,这正是系统辨识所渴求的特征;而基于免疫原理提出的各种优化算法是一种具有优越性能的全局优化算法,非常适合预测控制的滚动优化等各种优化问题的求解。论文主要研究内容及贡献如下:(1)提出一种抗体结构编码方法及基于结构编码的免疫优化算法,实现非线性系统结构辨识。该算法基于免疫系统原理,将抗体的非线性响应模型编码为动态结构树,通过结构树的克隆、选择、变异、交叉等免疫操作来实现非线性问题的免疫优化,实现了非线性系统模型的结构辨识。(2)提出一种抗体的混合编码方法及基于混合编码的免疫优化算法,应用于非线性系统模型的结构与参数的一体化辨识。混合编码方法将非线性表达式的结构通过动态结构树来描述,表达式的参数通过动态浮点数组来描述,代表非线性表达式的抗体的完整编码由动态结构树编码与浮点编码混合组成。该算法通过对结构编码与参数编码的免疫优化操作实现非线性表达式结构与参数的免疫优化,实现非线性系统模型的结构与参数的一体化辨识。基于结构编码与混合编码的免疫优化算法均具有卓越的全局搜索性能,不依赖过多的先验知识,应用于非线性系统辨识时,能得到结构简单、容易理解的非线性表达式模型。(3)提出一种基于克隆选择的免疫预测控制算法。采用比较通用的NARX形式的预测模型,通过克隆选择算法求解滚动优化问题,利用预测模型及目标函数在解空间中寻优直接获得预测时域内的最佳控制序列,避免了求解Diophantine方程与逆矩阵及复杂的推导过程。该算法对非线性系统不需要进行线性化,对带强耦合的MIMO系统不需要解耦,使用罚函数处理约束也非常方便。仿真结果表明基于克隆选择的免疫预测控制算法对外部干扰及建模误差具有很好的鲁棒性;并且不修改算法及算法参数就能对时滞系统、非最小相位系统、不稳定对象、非线性系统及MIMO系统等实现理想的控制效果,因而具有通用性,有利于预测控制的应用推广。(4)设计一种基于自适应免疫预测控制的智能控制仪表。该仪表将基于混合编码免疫优化的系统辨识算法与基于克隆选择的预测控制算法结合,采用基于多DSP并行实现的模块化硬件/软件结构,具有并行性、自适应性、鲁棒性、通用性及易于使用与维护等特点,有望成为优于传统PID仪表的新一代智能控制仪表。(5)针对船舶载重货物计量难的现状,研制了一种新型的船舶载重智能计量仪。该计量仪首次利用混合编码免疫辨识算法实现船舶载重模型的辨识,通过检测船体四个位置的吃水深度,根据辨识出的船舶重量模型计算船舶重量,从而实现装/卸货物的计量。工程实践表明,该船舶载重智能计量仪计量误差小于0.5%,完全满足船舶运输中对中低价货物的计量要求,同时为船舶载重在线测量提供了一种新型计量方法。更多还原

【Abstract】 Based on the immune principium, two significant fields concerning about nonlinear system control: the nonlinear system model identification and the predictive control are discussed in this dissertation. The immune system is emerging as a distributed dynamic robust control system with powerful self-study ability and more and more researchers attach importance to the artificial immune algorithm based on the immune principium, which is widely employed in numerous fields. Although with finite categories, the antibody in immune system is able to identify the antigen which is of multifarious categories and continuous evolution. The mechanism, which the antibody identifies the antigen, verifies that the immune system is of powerful self-study, self-organizing and favorable self-adaptive performances which are needed by normal system identification. Diversified optimization algorithms based on the immune principium are global optimization algorithms with excellent performances which are fit for diversified optimization problems such as rolling optimization in predictive control. The main research content of this thesis and contribution are below:(1) A coding method for antibody structure and an immune optimization algorithm based on the structure coding are proposed here, which are applied to nonlinear system structure identification. This algorithm is based on the principium of immune system and encodes the nonlinear response model of the antibody as a dynamic structure tree, which can be utilized in the immune optimization for nonlinear problems through the immune operations such as clone, selection, mutation and crossover on the structure tree. The structure identification for nonlinear system can be achieved with the algorithm.(2) A hybrid coding method for antibody and an immune optimization algorithm based on the hybrid coding method are proposed here which are employed in the incorporate identification for the structure and parameters of nonlinear system model. The structure of the nonlinear expression is encoded as a dynamic structure tree though hybrid coding method, in which the parameters of the expression are described by dynamic floating point array. The operations of immune optimization for the nonlinear expression structure and parameters are achieved through the encoding of structure and parameters, which are used for the realization of incorporate identification for the structure and parameters of nonlinear system model. There are excellent performances in immune optimization algorithm based on structure coding and hybrid coding, such as predominant global searching ability, not so reliant in getting too much transcendent knowledge and easy to get an intelligible nonlinear expression model with simple structure.(3) A predictive immune clone algorithm based on clone selection is proposed. A predictive model based on universal form of NARX is employed and the rolling optimization problem is solved by the clone selection algorithm, in which the optimum control sequence in time domain is directly achieved by using predictive model and the searching for the optimum of the target function in solution space so that it can avoid the complex process of solving Diophantine function, inverse matrix and so on. It is no need for this algorithm to linearize the nonlinear system and to decouple the strong-coupling MIMO system. It is also convenient to use the penalty function to solve the restriction. The simulation results also verify that the algorithm exhibits corking robustness against the external interference and modeling errors. Ideal control effect can be achieved without changes of the algorithm and its parameters in time-delay system, non-minimum phase system, unstable object, non-linear system and MIMO system. It is so that why this algorithm has generality and is easy to use, which is favorable for the application and generalization of predictive control.(4) A novel intelligent watercraft load meter is developed against the status quo that the watercraft load is hard to measure. It is the first time for the hybrid coding immune identification algorithm to identify the watercraft load model and actualize the measuring of loading and unloading cargo. The practical engineering application verifies that the error of the proposed watercraft load meter is less than 0.5% which satisfies the requirement of measuring low price cargo in watercraft transportation completely and a novel method for watercraft load online testing is proposed at the same time.(5) An intelligent control meter based on adaptive immune predictive control is designed. This intelligent meter incorporates the system identification algorithm based on hybrid immune optimization and predictive control algorithm based on clone selection, whose hardware and software structure are actualized by multiple parallel DSP. It includes performances such as parallelism, adaptability, robustness, generality and easy for using and maintenance. So it will be a new generation of intelligent control meter which is more excellent than traditional PID meters.更多还原