基于稀疏存储Elman神经网络的直线伺服控制

Linear Servo System Control Based on Improved Elman Neural Network with Sparse Memory

为提高数控机床直线进给系统的动态跟踪性能及抗干扰能力,结合进给系统重复运动的特点,利用前一次或前几次的历史控制信息提高进给系统的动态性能,提出了具有动态稀疏存储功能的改进Elman神经网络;引入迅速联想的表格查询方式对神经网络的历史信息进行分类存储、选择利用以增强网络泛化能力,提高网络收敛速度;详细推导了改进Elman神经网络的数学模型及权值调整算法,并将其应用到直线进给伺服系统中,结果表明,基于稀疏存储Elman神经网络的速度控制器具有良好的跟踪精度和抗干扰能力。

In order to improve the dynamic performance and disturbance resistance abilities of CNC machine tool direct-drive linear servo system with complexities of dynamics and nonlinear, taking advantages of historical control information of repeated motion, an improved Elman neural network with spares memory was proposed and rapid associate theory based on table-look up was introduced to enhance learning speed and generalization capability of neural network. The information of neural network was classified, stored and selected to use. The mathematical model of improved Elman neural network and the weight adjustment algorithm were derived in detail and applied to a linear feed servo system.The results show that the controller based on improved Elman neural network with spares memory exhibits satisfactory performance in tracking precision and disturbance resistance.