Modeling of unsteady heat conduction field by using composite recurrent neural networks
Yasuaki Kuroe, Ichirô KIMURA · 2002
This paper presents a method for modeling a class of distributed parameter systems, unsteady heat conduction fields, by using neural networks. A new architecture of recurrent neural networks in which the dynamic and static neurons are arbitrarily connected is introduced and their training algorithm is derived. A synthesis procedure for determining structures of the composite recurrent neural networks is derived from the qualitative knowledge on the dynamics of unsteady heat conduction fields. It is shown through numerical experiments that the proposed method can realize suitable models of unsteady heat conduction fields on the networks.