A special type of recurrent neural networks is the Elman network. Feedforward connections of the Elman network can be trained essentially as feedforward networks by means of the simple backpropagation algorithm, their feedback connections have to be kept constant. It is important to select correct values for the feedback connections for the training to convergence. However, finding these values can be a lengthy trial-and-error process. This paper describes the use of simulated annealing (SA) algorithm to train the Elman network for dynamic systems identification. The SA algorithm is an efficient random search procedure which can simultaneously obtain the optimal weight values of all connections.
Subjects | Electrical Engineering |
---|---|
Journal Section | Research Articles |
Authors | |
Publication Date | December 31, 2002 |
Acceptance Date | July 1, 2002 |
Published in Issue | Year 2002 Volume: 15 Issue: 2 |