Abstract: A state-dependent routing algorithm based on Hopfield's neural networks model is given, which takes advantage of other dynamic routing methods for the circuitswitched network. The model fully integrates the traffic data, network states and circuits reservation policy monitored by the network management center, and optimizes dynamic routes for the whole network on the bases of short period forecasting of trunk congestion probabilities. The design of the model is ascloser to the real world as possible, and also takes the hardware realization into consideration. Computer simulations show that the algorithm is much better than the classical fixed routing methods, therefore the model could be a solutionfor the applications of neural networks to the dynamic routing.

Key words: neural networks, optimization algorithm, dynamic routing