摘要
建立了一种采用改进的自适应遗传算法实现动态递归的进化E lman神经网络模型。提出了对网络的结构、权重、结构单元的初始输入和自反馈增益因子同时进化的学习算法。用初始状态优化的E lman网络集成反馈学习算法和E lman网络在线训练两种动态辨识算法形成的集成化动态递归网络辨识算法,实现了超声马达的速度辨识。模拟结果表明,提出的算法不仅实现了动态递归网络的全自动优化设计,而且明显提高了动态递归网络模型辨识算法的收敛精度,为非线性系统辨识提供了一条新的途径。
Using an improved adaptive genetic algorithm, an evolutionary Elman neural network model with dynamic regression was built. A learning algorithm which can simultaneously evolve the network structure, the weights, the initial inputs of structural unit, and the self-feedback gain coefficient was proposed. Ultrasonic motor speed identification was performed by the dynamic recurrent network algorithm which was formed by the integration between initial state optimized Elman network with feedback learning algorithm and Elman network on-line training algorithm. Simulated results show that the proposed algorithms not only realize fully automatic optimization design for the dynamic recurrent neural network, but also improve convergent accuracy for the model identification. And this method can provide a new approach to nonlinear dynamic system identification.
出处
《吉林大学学报(工学版)》
EI
CAS
CSCD
北大核心
2005年第5期511-519,共9页
Journal of Jilin University:Engineering and Technology Edition
基金
国家自然科学基金资助项目(19872027)
高等学校博士学科点专项科研基金资助项目(20030183060)
吉林省科技发展计划项目(20030520)
教育部科学技术研究重点项目(02090).
关键词
计算机系统结构
控制理论
计算智能
动态递归神经网络
非线性系统辨识
computer system structure
control theory
computing intelligence
dynamic recurrent neural network
non-linear system identification