履歴データを事例として使用する非線形モデリング技術 TCBM: Topological Case Based Modeling

抄録

Current system identifications require nonlinear modeling technology. The system to be modeled has been effected by disturbances and has been related to human nonlinear actions. For this purpose, the black box modeling technology such as neural network is useful. However, the black box models can not show the qualitative relation between inputs and output. Therefore, the black box models require complete data set for modeling. In the real world, the complete data set seldom be obtained from the actual system. In spite of the data are incomplete, the traditional black box modeling techniques convert the information of the incomplete historical data to the model parameters which regulate the relations between input space and output space. These specified model parameters can not evaluate the reliability of the output for new input situations. This is a barrier to the actual use of traditional black box model for the real world system.
In this paper the nonlinear modeling technology using historical data for case is proposed. This technology is named TCBM (Topological Case Based Modeling). TCBM employs case based reasoning (CBR) technology for modeling to cope with the problems of traditional modeling technology that is a reliability of a estimated output on condition that the incomplete system data are used for modeling. CBR theory provides no general method to define the similarity between cases. In contrast, TCBM can generally define the similarity on the condition that the system to be modeled satisfies the continuous relationship between inputs and output. As a result, TCBM can easily construct its model same as traditional black box modeling technology and can show the reliability of the estimated output.