Distributed-information neural control: The case of dynamic routing in traffic networks

Large-scale traffic networks (e.g., computer and communication networks, fr eeway systems, etc.) can be modeled as graphs in which a set of nodes (with storing capacities) are connected through a set of links (where traffic de lays and transport costs may be incurred) that cannot be loaded above their traffic capacities. Traffic flows may vary over time. Then the nodes (i.e. , the decision makers acting at the nodes) may be requested to modify the t raffic flows to be sent to their neighboring nodes. In this case, a dynamic routing problem arises. The decision makers are realistically assumed 1) t o generate their routing decisions on the basis of local information and po ssibly of some data received from other nodes, typically, the neighboring o nes and 2) to cooperate on the accomplishment of a common goal, that is, th e minimization of the total traffic cost. Therefore, they can be regarded a s the cooperating members of informationally distributed organizations, whi ch, in control engineering and economics, are called team organizations. Te am optimal control problems cannot be solved analytically unless special as sumptions on the team model are verified. In general, this is not the case with traffic networks. An approximate resolutive method is then proposed, i n which each decision maker is assigned a fixed-structure routing function where some parameters have to be optimized. Among the various possible fixe d-structure functions, feedforward neural networks have been chosen for the ir powerful approximation capabilities. The routing functions can also be c omputed (or adapted) locally at each node. Concerning traffic networks, we focus attention on store-and-forward packet switching networks, which exhib it the essential peculiarities and difficulties of other traffic networks. Simulations performed on complex communication networks point out the effec tiveness of the proposed method.