FHOPFIELD ANN MODELS FOR ROUTING IN COMMUNICATION NETWORKS - A SIMULATION STUDY

Abstract

Neural network applications have experienced a spurt in recent years. Inspired by the neural systems, the neural network models are parallely connected networks. The property of massive parallelism makes these models an ideal too] for computational work in constrained optimization problems. This capability could be exploited to solve the optimum data traffic routing problems in the communication networks. This report embarks on investigating the properties of some of the neural network algorithms for routing, available in literature so far. Starting with a reviewal survey of artificial neural network, this report focuses on the use of Hopfield networks in solving routing problems. Initially the minimization process is implemented using a modified version of neural network travelling salesman algorithm. Then a modified version of above, with the capability to handle dependent component failures is studied. Finally a shortest path algorithm, which is adaptive and could be implemented in real time, has been considered. The computer simulation results have been included to illustrate the effectiveness of various approaches.