A Class of Rough GMDH for Uncertain Complex System Modeling and Positive Analysis
一类基于不定性复杂系统的粗糙GMDH模型及实证分析

The notion of rough set is leaded into GMDH(Group Method of Data Handling) for the first time. At the same time, the estimation method for coefficients in each model of each level of front-propagation network is turned from traditional regression into uncertain programming and linear programming. By means of this transform, such rough GMDH(Rought Group Method of Data Handling) is proposed that to increase the analyzing ability of modeling and mid-term or long-run prediction ability. By means of DM's subjective experience, the information needed by modeling can be complete, which may decrease the demand for sample set size, even may decrease to the size with only one sample. In addition, for mining the DM's subjective preference and achieving man-machine interaction, one man-machine interactive system with trial mechanism is designed to provide the DM more complete information of results, which may provide a good decision support for the DM to reduce the difficulty when meeting in three-fold problem including accuracy, information redundancy and scarcity. Finally, a simple example researching the relation between company core competence and economic performance is proposed to explain RGMDH modeling.