一般二维模糊控制器的等效神经网络建模与验证  

Modeling and verification for equivalent NN of general two-dimensional fuzzy controller

作　　者：董淳[1] 杨绍荣 李太福[3] 苏盈盈[1] 

机构地区：[1]重庆工学院电子信息与自动化学院,重庆400050 [2]重庆三峡水利电力(集团)股份有限公司,重庆400050 [3]重庆科技学院电子信息工程学院,重庆400050

出　　处：《计算机工程与设计》2008年第11期2912-2914,2917,共4页Computer Engineering and Design

基　　金：重庆市自然科学基金项目(KJ071411)。

摘　　要：针对模糊控制器的计算复杂性,实时性能差,易产生维度灾难等问题,利用神经网络的万能函数逼近能力,构建一个神经网络模型,精确的逼近已知的模糊控制器,从而减少运算量,实现实时控制。以一个已知的二输入单输出模糊控制器为例,建立一个与之等效的神经网络,通过训练,使得精确的逼近模糊控制系统。最后,给定输入信号,分别用模糊控制器和神经网络控制同一个被控对象。结果表明,用一个与模糊控制器等效的神经网络来控制同一个对象,控制效果非常相似。因此,用模糊控制器的等效神经网络模型代替,在实现环节上可以减少计算复杂性,维度灾难,提高实时性能。Aimed at implementing puzzles of fuzzy controller, such as computational complexity, dimensional disaster and requirement of real-time control, the universal approximation characteristic of neural networks is utilized to accurately approximate a known fuzzy controller with an equivalent NN. At first, modeling an equivalent NN with two-input and single-output fuzzy controller, it could ac- curately approximate fuzzy control system via data pairs training. Furthermore, the same process model is controlled by the fuzzy controller and the equivalent NN with different reference inputs, respectively. Results show that control qualities from two different controllers are extremely similar. Therefore, the fuzzy controller can be replaced by a reconstructing NN model in purpose of reducing the computational complexity, dimensional disaster and improving the real-time performance.

关 键 词：模糊控制器 神经网络 函数逼近 建模 维度灾难 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]