| 초록 |
Artificial Neural Networks (ANNs) have been widely used to describe nonlinear dynamics due to their excellent performances. But, most previous ANNs have been developed for discrete-time system identification. As a result, even though they can model discrete-time systems with fairly good accuracy, their model performances can be seriously poor when the sampling time is small and/or the process is close to a continuous-time dynamic system. In this paper, we confirm that the disadvantages of the previous approaches experimentally. And, we develop a new continuous-time recurrent neural network model to overcome the problems. The superiority of the proposed approach compared to the previous ones is demonstrated experimentally by applying them to identify the nonlinear dynamics of a micro-PCR reactor system. |
