摘要
针对多变量系统解耦控制的要求和特点,传统的PID神经网络在选取初始权值难以确定,往往是随机得到,容易导致采用的BP学习算法陷入局部极值。提出了一种人工鱼群算法优化PID神经网络初始权值。通过对多变量控制对象的matlab仿真验证,把人工鱼群算法优化得到的最优初始权值带入PID神经网络,结果显示加快了PID神经网络的收敛速度,使控制量迅速地接近控制目标,保证了系统稳定性,取得了满意的控制效果。
As to the requirements and characteristics of decoupling control of multivariable system, the traditional PID neural network can not easily decide the initial weights. Rather it gets the weights randomly, often leading to lo- cal extremum by the adopted BP learning algorithm. This paper proposed an artificial fish -swarm algorithm to opti- mize PID neural network initial weights. Through the Matlab simulation experiment for multivariable control objects, the optimal initial weights gained by optimizing artificial fish - swarm algorithm were brought into the PID neural net- work. It accelerates the converging of PID neural network, makes the control variables closer to control objectives, improves the system stability and achieves satisfactory control effect.
出处
《计算机仿真》
CSCD
北大核心
2014年第10期350-353,共4页
Computer Simulation
基金
内蒙古自治区高等学校科学研究项目(NJZY13103)
内蒙古自然科学基金项目(2013MS0919)
内蒙古工业大学重点科学研究项目(ZD201235)
关键词
人工鱼群算法
神经网络
初始权值
多变量
解耦控制
Artificial fish - swarm algorithm
Neural network
Initial weight value
Muhivariable
Decoupling control
