云神经网络的函数逼近能力  

Function approximation capabilities of cloud reasoning neural networks

作　　者：于少伟[1] 

机构地区：[1]山东英才学院商学院,山东济南250104

出　　处：《山东大学学报（工学版）》2010年第4期62-66,共5页Journal of Shandong University（Engineering Science）

基　　金：山东省自然科学基金资助项目(ZR2009GQ016);山东省高等学校科技计划资助项目(J09LG68)

摘　　要：运用云理论,建立自适应神经-云推理系统(adaptive neuro-cloud inference system,ANCIS)的控制模型,并证明该模型具有全局逼近性质。首先将Zadeh模糊推理神经网络变为云推理网络,建立一个多输入单输出的T-S型ANCIS模型;然后设计系统变量的属性函数和推理规则,确定输入输出关系以及系统输出结果的计算表达式;最后通过证明所建模型的输出结果计算式满足Stone-Weirstrass定理的3个假设条件,完成该模型的全局逼近性证明。A controlling model of an adaptive neuro-cloud inference system（ANCIS） was constructed by utilizing cloud theory with the function approximation property of the proved model.The Zadeh fuzzy inference neural nets were developed into cloud inference nets,and a model of ANCIS in Takagi-Sugeno type was established.The attribute functions and the inference rules of the system variables were devised with computational relations between the layers of input and output and a synthesized computational expression of system outputs ascertained.A successful proof of global approximation property of the model was accomplished by testifying that the computational expression of output results of the founded model satisfied the three hypothetic conditions of the Stone-Weirstrass theorem.

关 键 词：云理论 云推理 神经网络 函数逼近 

分 类 号：TP182[自动化与计算机技术—控制理论与控制工程]