Utilisation des réseaux de neurones temporels pour le pronostic et la surveillance dynamique: Etude comparative de trois réseaux de neurones récurrents

This article gives a state of the art of temporal neural networks and a comparison of three recurrent neural network which are most representative for applications of dynamic monitoring and prognosis. The criteria of selection of these networks are at two levels: a temporal criterion and an architectural criterion. Following the application of these criteria, three recurrent networks seem relevant: the RRBF, the R2BF and the DGNN. Tests using a benchmark of dynamic monitoring and a benchmark of prognosis enable us to evaluate the performances of the three temporal networks in term of computing and processing capacity time.