非线性量化小脑模型关节控制器神经网络控制器

Nonlinear quantization cerebella model articulation controller neural network controller

中文关键词  小脑模型神经网络；非线性量化；分段量化；神经网络控制器  神经网络  学习算法

英文关键词  cerebella model articulation controller (CMAC)  nonlinear quantization  piecewise quantization  neural network controller  neural networks  learning algorithms

中文摘要

      常规CMAC神经网络采用线性均匀量化, 稳态控制精度与量化级数相关, 增加量化级数可提高稳态精度但会导致内存空间和计算量的增加. 本文提出一种可采用幂函数、高斯、分段三种非线性量化方法的非线性量CMAC神经网络, 并分析了非线性量化CMAC的收敛性, 解释了非线性量化提高稳态精度高的本质. 面向一阶惯性环节、二阶系统、一阶时变系统及二阶时变系统, 分别跟踪方波、斜坡、正弦波、三角波和加速度等输入信号, 仿真验证了非线性量化CMAC神经网络控制器的有效性, 给出了不同非线性量化方法的适用性. 结果表明, 非线性量化CMAC参数容易设定, 物理意义清晰, 与常规CMAC对比, 其快速性和控制精度显著提高, 可以有效解决实际复杂非线性时变系统的控制.

英文摘要

      Equal-size quantization method is adopted in the conventional CMAC (Cerebella Model Articulation Controller) neural network. The size of quantization series plays an important role in the steady-state control accuracy. Large quantization sizes may take up too much storage memory and make calculations complex while small quantization sizes will decrease the control accuracy of CMAC. A nonlinear quantitative CMAC neural network is proposed in this paper, which uses the power function, Gaussian function and piecewise function to quantify. The convergence of the nonlinear quantitative CMAC is derived. Meanwhile the reason for that the nonlinear quantization can improve the accuracy of CMAC is explained by analyzing the changes of activation memory unit under different quantization methods. Finally, four different plants are selected as researching objects, such as first-order inertia link, a second order system, a first time varying system and a second order time-varying system. Simulations on controlling these four different plants are given to verify the effectiveness of the nonlinear CMAC, tracking the square wave function, slope function, sine wave function, triangular wave function and acceleration function respectively. Compared with CMAC controller, the stable accuracy and rapidity of nonlinear quantitative CMAC controller is better than that of CMAC. From the simulation result, it can be seen that nonlinear quantized CMAC controller parameters are easy to set, which is suitable for the complex and nonlinear time-varying control system.