基于遗传算法的连续系统模型辨识

来源期刊：中南大学学报(自然科学版)2001年第3期

论文作者：贺尚红 唐华平 段吉安 钟掘

文章页码：305 - 308

关键词：系统辨识;遗传算法;参数估计;优化

Key words：system identification; genetic algorithm; parameter estimation; optimum

摘    要：系统辨识是控制系统设计的基础,传统的系统辨识大都是对离散数学模型的参数估计,发展了以最小二乘法为基础的理论和方法.而连续模型则需经过离散模型的转换方可得到.作者基于遗传算法和系统仿真技术,建立了输出端噪声存在的连续系统模型辨识方法.遗传算法是建立在自然遗传学机理上的参数搜索方法,但常规的遗传算法具有收敛速度慢,因封闭竞争易导致“早熟”等缺点,为此,通过引入自适应适值函数和动态变异因子,采用精英保护策略,加速了寻优过程,改善了遗传算法的全局搜索及局部寻优能力,为连续系统模型辨识提供了新的途径.

Abstract: Identification of system is the foundation of control system. In most cases, control systems are modeled by means of discrete mathematical models. And the parameters are estimated by least square based algorithms. The discrete models can be transformed into continuous system models, but it may lead to changing the stability of the systems. Based on genetic algorithm and system simulation technique, a new identification method of continuous system is established.Genetic algorithm is an optimal searching approach based on natural selection and natural genetic mechanism. But the conventional genetic algorithmhas drawbacks such as lowsearching efficiency and premature convergence. The paper introduced adaptive fitness function and mutation rate, and the“elitism”selection approach is adopted, so the performance of convergence to the global optimum is improved, and the occurrence of premature converge is avoided. The validity and efficiency of proposed algorithm are illustrated by simulated results.