2017,53(9)
1 引言
随着无线通信技术的发展越来越迅速,其对通信系
统性能的要求也随之提高。作为无线射频前端的重要
部件之一,射频功率放大器(Power Amplifier,PA)也因
此得到了越来越多的关注。RF 功率放大器在大信号激
励下会表现出强非线性与记忆效应
[1-2]
,从而增加了 PA
建模的复杂度。因此,如何提高具有非线性记忆功放的
建模精度,使其在系统级仿真中能获得更优的性能,成
⦾理论与研发⦾
基于分组混沌 PSO 算法的模糊神经网络建模研究
张 楠,南敬昌,高明明
ZHANG Nan, NAN Jin gchang, GAO Mingming
辽宁工程技术大学 电子与信息工程学院,辽宁 葫芦岛 125105
School of Electronic and Informa tion Engineering, Liaoning Technical University, Huludao, Liaoning 125105, China
ZHANG Nan, NA N Jingchang, GAO Mingming. Fuzzy neural networ k for ampl ifier power modeling based on
grouping parallel-chaotic Particle Swarm Optimization. Com puter Engineering and Applications, 2017, 53(9):31-37.
Ab stract: In order to improve the accuracy of radio frequenc y power amplifi er with memory eff ect, and th e early fas t
convergence rate of the traditional particle swarm optimization algorithm, but in the later period easy to fall into premature
and local optimum characteristic s, a group of parallel chaotic particle swarm optimization algorithm is proposed a nd the
dynamic fuzzy ne ural net work param eters ar e optimized by using the algorithm to optimize the dynamic fuzzy neu ral
net work param eter s. The grou ping parallel chaotic particle s war m optim ization a lgorithm is used to combine the grouping
method and chaotic particle swarm optimization algorithm. The populat ion can be divided into several gr oups. Each group
computes independently t o impr ove t he con vergence rate , while the cha os theory is applied to e ach particle to avoid
premature and local optimum, shortening the ti me of iteration. By the simulation, the training err or of the model is
reduced to 0.1, and the convergence rate is improv ed by 32. 5%, which verifies the validity and reliability of the method.
Key words: chaos theory; groupin g parallel partic le swarm optimization; dynamic fuzzy neural network; memory power
amplifier model
摘 要:为改善记忆功放建模的精度,且针对粒子群算法早期收敛速度较快,但在后期易陷入早熟收敛,局部最优
等特点 ,提出了一种分组并行混沌粒子群优化算法(Grouping Parallel-Chaotic Pa rticle Swarm Optimization,GP-
CPSO),将分组粒子群优化算法与混沌思想相结合 ,并用该算法优化动态模糊神经网络(Dynamic Fuzzy Neural
Network,DF NN)参数,建立 DFNN 功放模型。引入分组的 CPSO 群算法,将种群划分为若干个组,每组单独计算,大
大提高了收敛速度,同时将混沌思想运用到每个粒子当中去,避免早熟和局部最优,缩短了迭代时间。通过仿真结
果可以看到,GP-CPSO 优化后的动态模糊神经网络建模的训练误差减小到 0.1 以内,收敛速度提高 32.5%,从而验证
了这种建模方法有效且可靠。
关键词:混沌思想;分组并行粒子群算法 ;动态模糊神经网络;记忆功放模型
文献标志码:A 中图分类号:TP301.6 doi:10.3778/j.issn.100 2-8331.1510-0259
基金项目:国家自然科学基金(No.61372058);辽宁省高等优秀人才支持计划项目(N o.LR2013012);辽宁省教育厅科学研究一般
项目(No.L2015209)。
作者简介:张楠(1990—),女,满族,硕士研究生,主要研究方向为射频功放神经网络建模,E-mail:zhangnan126mail@12 6.com;南
敬昌(1971—),男,博士,教授,硕士生导师,主要研究方向为射频电路与器件,多媒体信息编码,通信系统仿真等;高明
明(1980—),女,蒙古族,博士,副教授,硕士生导师,主要研究方向为无线通信、射频通信等。
收稿日期:2015-10-27 修回日期:2015-12-15 文章编号:1002-8331(2017)09-0031-07
CN KI 网络优先出版:2015-12-31, http://www. cnki.net/kcms/detail /11.2127.TP. 20151231.1148.020. html
Computer Engineering and Applications 计算机工程与应用
31