|
In the fuzzy linear regression model , studies were focus on the output and parameters as fuzzy numbers, and the input as nonfuzzy real numbers. One ofthese fuzzy linear regression analyses is Tanaka linear programming method and the other is the least square method. Tanaka method is to consider the fuzziness of models and then transform the method of parameter estimate into linear programming method. Least square method is to define the distance between two fuzzy numbers, and then define a least square objective function. Moreover, the concept of noise cluster is used to yield a robust fuzzy least square method. In this thesis, we regard the output , parameters and the inputs as fuzzy numbers. Although Sakawa and Yano (1992) give a new Tanaka linear programming method to estimate the parameters of this model, no one investigates the least square approach in this complicated model. This is the subject of the thesis. Since the product of two LR type fuzzy are not always the corresponding LR type. Therefore, we first use the interval to represent the product of fuzzy numbers and apply the least square method to get a least square estimate, which is called interval least-square estimate. On the other hand, we use the approximate introduced by Dubois and Prade (1980) and the distance defined by Yang and Ko (1997) to get the fuzzy least square method.
|