重卷机组张力神经网络控制器的可视化构建

Visualization building of neural network controller for recoiling tension control system

针对重卷机组中张力控制系统存在参数时变、非线性等问题,引起张力控制模型仿真的复杂性,为简化实际应用中神经网络控制器构建的复杂性,本文结合Matlab、VC++工具建立了一种可视化调节界面用以获得神经网络控制器。通过实际输出曲线同期望轨迹的误差对控制输入进行可视化调节,利用学习控制对控制输入做更进一步的优化,进而训练得到期望的神经网络控制器。为克服单纯采用神经网络控制造成的稳态误差,设计了切换策略用以实现不同偏差作用下神经网络控制器同PI控制器之间的切换。通过仿真验证了该方法的有效性。

The problems of parameter time-variable and nonlinear exist in the tension control system of recoiling unit,which makes the tension control model simulation complicated. To simplify building neural network controller in real application,a visual adjustment interface is set up to the NN controller based on Matlab and VC ＋＋. Through the error between actual output curves and expected trajectory,control input is adjusted in a graphical user interface（ GUI）,besides,control input is further optimized by learning control concept,finally,the NN controller is trained and obtain the expected one. The switch between the NN controller and PI controller is designed under the different deviation,and it can overcome steady state error come from using only the NN controller. The proposed method is verified effective by simulation.