メンバーシップ関数の微分に関する一考察(<一般演題>診断システム)

抄録

The purpose of this work is to consider a method which minimizes the affects of input data error on the output of a decision making system. Theinput affects are considered to be proportional to the differential of amembership function. Such membership functions as the threshold function, the ramp function, the standard membership function and the integral of the normaldistribution function as an approximation of an error distridution function ofthe input data, ware compared. Two membership functions which minimize theinput affects are proposed : an integral of the normal distribution function having the same standard deviation to the error distribution function, and an approximation of this function, a ramp function with a width as large as 4SD of the error distribution function.