改进蚁群神经网络及其在滞后系统中的应用

An Improved Ant Colony Neural Network and Its Application to Time Lag Systems

针对热连轧监控AGC系统具有多变量、强耦合、非线性、大滞后及参数时变的特点,提出了一种改进蚁群神经网络,采用动态局部信息素更新和自适应调节信息素挥发的全局信息素更新相结合的方式对蚁群算法进行了改进。并利用改进后的蚁群算法对神经网络权值和阈值进行优化,利用优化后的神经网络对PID控制器参数进行整定。改进的蚁群算法稳定性好,寻优效率高,避免了神经网络参数陷入局部极小等问题,从而实现大滞后系统的优化控制。仿真结果表明,在监控AGC系统中,在对象的滞后时间发生变化的情况下,基于改进蚁群神经网络的优化控制系统具有动态响应速度较快,对外部扰动具有良好的鲁棒性,使控制品质得到很大的提高。

Aiming at the characters of feedback AGC such as muhivariable, strong coupling, nonlinearity, time lag and time-varying parameters in hot rolling, an improved ant colony neural network is developed. The dynamic local pheromone updating rule and the adaptive global pheromone updating rule are proposed in the improved ant colony arithmetic and used for tuning weight parameters of neural network. The neural network is used for adjustment of PID parameters to realize time lag optimal control. The new control algorithm has a good stability and an efficient searching speed. It avoids the problem of controller parameters trapping into local minimum. Simulation results show that, under different time lag, the system also has many advantages such as fast response, strong robustness and good self-adaptability.