基于改进差分进化算法的系统辨识方法  

System identification method based on improved differentialevolution algorithm

作　　者：吕证 陈博[2] 李章勇[2,3] 秦对 LYU Zheng;CHEN Bo;LI Zhangyong;QIN Dui(School of Automation,Chongqing University of Posts and Telecommunications,Chongqing 400065,P.R.China;School of Life Health Information Science and Engineering,Chongqing University of Posts and Telecommunications,Chongqing 400065,P.R.China;School of Optoelectronic Engineering,Chongqing University of Posts and Telecommunications,Chongqing 400065,P.R.China)

机构地区：[1]重庆邮电大学自动化学院,重庆400065 [2]重庆邮电大学生命健康信息科学与工程学院,重庆400065 [3]重庆邮电大学光电工程学院,重庆400065

出　　处：《重庆邮电大学学报(自然科学版)》2025年第1期37-45,共9页Journal of Chongqing University of Posts and Telecommunications(Natural Science Edition)

基　　金：国家自然科学基金项目(11904042);中国博士后科学基金项目(2022MD723728);重庆市自然科学基金项目(CSTB2023NSCQ-MSX0861)。

摘　　要：对于复杂控制系统,采用传统建模方法得到的被控对象模型,存在精度较低和收敛速度较慢等问题。针对该问题,提出了一种基于改进差分进化算法的系统辨识方法,通过采用变异率和交叉算子动态更新策略,并实时更新种群以及优化、进化终止条件,大幅度地提升了辨识模型的精度以及算法的收敛速度。与传统的最小二乘法、二阶加时滞模型法、差分进化算法以及遗传算法相比,提出的辨识方法在精度上分别提升了94.91%、40.11%、23.33%和8.48%。与传统遗传算法和差分进化算法相比,提出的方法在收敛速度上提升了近3.2倍。实验结果表明,提出的方法可以有效地克服干扰信号影响,得到精确的系统传递函数模型。For complex control systems,models obtained using traditional modeling methods often suffer from low accuracy and slow convergence.To address this issue,a system identification method based on an improved differential evolution(DE)algorithm is proposed.The method incorporates dynamic update strategies for mutation rates and crossover operators,real-time population updates,and optimized evolution termination conditions,significantly enhancing model accuracy and algorithm convergence speed.Compared to traditional least squares,second-order plus time-delay models,DE algorithms,and genetic algorithms,the proposed method improves accuracy by 94.91%,40.11%,23.33%,and 8.48%,respectively.Additionally,it achieves a nearly 3.2 times improvement in convergence speed over conventional genetic and DE algorithms.Experimental results demonstrate that the proposed method effectively mitigates the impact of interference signals and produces precise system transfer function models.

关 键 词：系统辨识 改进差分进化算法 超声换能器 传递函数模型 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]