An expert system for network control problems and its applications in large scale network design under uncertainty

This paper describes an expert system to find the control parameters in order to optimize the performance function of network when the network users are intelligent and we have granular information about network. Because of intelligent agents, a traffic assignment model is applied to predict the link flows. The uncertainty of granular information of network is also captured with fuzzy sets. We obtain equilibrium flows from a traffic assignment model with fuzzy costs. The performance function is concluded from a simulation scheme considering three criteria: network congestion, traveling time and social attachment. To optimize the performance function with respect to control variables, we use particle swarm optimization (PSO) approach. Through optimization process in the case of large scale networks, a lot of evaluation of the performance function is necessary which is computationally heavy. Thus a multilayer perceptron is used as a metamodel to predict the system behavior when the control parameters are varying. Both of the components of the proposed expert system, metamodel and the optimizer can be implemented in parallel manner, thus it is possible to find near optimal control parameters in large scale networks. Such method can be pursued to deduce the congestion through urban network when the links are controlled by cost instruments using RFID technology or camera with signal processing. The place of such algorithm in network design is also investigated.