This paper is concerned with the optimum routing and flow assignment problem in packet-switch networks. The optimum objective function is the network-wide average time delay. To make the solution be implemented reliably in real time, a neural network for shortest path computation that is a two-layer recurrent structure is applied to flow deviation method. Simulation results show that improvements can be achieved in the reliability of successful convergence and in the decrease of computation time.
Ali M Kamount F. Neural networks for shortest path computation and routing in computer networks[J].IEEE Trans. on Neural Networks.1993,4(6):941-953[2]Fratta L, Gerla M and Kleinrock L. The flow deviation method: An approach to store-and-forward communication network design[J].Networks.1973, 3(1):97-133[3]Schwartz M and Cheung C. The gradient projection algorithm for multiple routing in message switched systems. IEEE Trans. on Communications. 1976, Com-24 (4): 449-456.[4]Filip A, Bernardete R, Luis R. A neural network for shortest path computation[J].IEEE Trans. on Neural Networks.2001, 12(5):1067-1073[5]Strum R D, Ward J R. Electric Circuits and Networks (2nd edition). Englewood Cliffs, New Jersey: Prentice-Hall Inc, 1985, 345-358.[6]Hopfield J J and Tank D W. Neural computation of decisions in optimization problems. Biolog. Cybern., 1985, 52(3): 141-152.[7]Wang J. A recurrent neural network for solving the shortest path problem[J].IEEE Trans. on Circuits and Systems _ I: Fundamental Theory and Applications.1996, 43(6):482-486[8]Bu L and Chiueh T. Solving the shortest path problem using an analog network[J].IEEE Trans. on Circuits and Systems _ I: Fundamental Theory and Applications.1999, 46(11):1360-1363