摘要
人工神经网络的训练问题实质上是一个优化问题。将模拟退火算法和基本粒子微粒群算法相结合,提出一种基于模拟退火的微粒群算法,该算法能够有效抑制早熟收敛。利用基于模拟退火微粒群算法优化BP神经网络的权值和阀值,有效的解决了BP算法易陷入局部极小值的缺点,从而提高了神经网络的精度和收敛速度。通过对非线性系统进行Matlab仿真研究,实验结果表明,基于模拟退火的微粒群算法训练的神经网络是一种有效的辨识方法。
The problem of neural network learning is a function optimization problem. An optimization algorithm--PSOSA （PSO with simulated annealing） which combine particle swarm optimization （PSO） and the simulated annealing （SA） is proposed, and can alleviate the premature convergence validly. Through optimizing the parameters of BP neural network, PSOSA algorithm is effective to solve the defect that other BP algorithms easily plunge into local solution, and can achieve convergence faster and precision higher. The simulation of nonlinear system based on PSOSA-BP neural network is performed using Matlab. The result shows that PSOSA-BP neural network is an effective identification method of nonlinear system.
出处
《计算机工程与设计》
CSCD
北大核心
2011年第2期667-670,共4页
Computer Engineering and Design
基金
云南省教育厅科研基金项目(2010Y060)
关键词
模拟退火
神经网络
粒子微粒群算法
辨识
训练算法
simulated annealing
neural network
particle swarm optimization （PSO）
identification
training algorithm
