摘要
充分利用PID结构简单、稳定性强的良好性能以及神经网络的自学习和自适应的特长,引入粒子群优化(PSO)学习算法,设计一种多变量自适应PID型神经网络控制器。神经网络的隐含层由带有输出反馈和激活反馈的混合局部连接递归网络组成,采用PSO学习算法优化神经网络参数。在深入研究分析PSO算法的基础上,引入变异因子和惯性权重自适应策略对该算法进行改进,既发挥了PSO算法随机优化收敛速度快的优点,又克服了该算法易陷入局部最优点的缺点,显著提高了控制系统的性能指标。最后,通过对二级倒立摆控制的仿真分析,证明该算法具有较强的鲁棒性。
Making full use of the advantages of simple construction and strong stability of PID and self-learning and adaptability of neural networks,introducing PSO(Particle Swarm Optimization) algorithm,a multivariable adaptive neural network controller was designed.It was composed of a hybrid locally connected recurrent network with an activation feedback and an output feedback in the hidden layer,PSO algorithm was applied in neural network parameter learning.On the basis of the research and analysis for PSO,the variability factor and inertia weight self-adaptability strategy were introduced to improve the PSO algorithm.The improved PSO algorithm not only plays the advantage of fast convergence speed,but also conquers the shortcoming of PSO algorithm which is easily plunging into the local minimum,and improved the performance of control system greatly.Finally,simulation of the control of double inverted pendulum shows that the proposed algorithm has strong robust ability.
出处
《系统仿真学报》
CAS
CSCD
北大核心
2011年第2期363-366,385,共5页
Journal of System Simulation
基金
国家自然科学基金(60774059)
关键词
多变量系统
PID型神经网络
PSO算法
二级倒立摆
multivariable
PID-like neural network
particle swarm optimization
double inverted pendulum