In this paper, we consider a traffic flow model where the information about the actual travel time for each alternative route is not available when each driver performs a route selection. For such a traffic flow model, we examine the effectiveness of two routing methods for minimizing the average travel time over all vehicles running in the model. One method is to minimize the average travel time globally. In this algorithm, a central manager determines the routes of all vehicles. Since the number of combinations of possible routes of drivers exponentially increases as the number of drivers, we need an efficient combinatorial optimization technique. In this paper, we employ a genetic algorithm to search for a near-optimal route choice of each driver. The other method is to minimize the average travel time locally by each driver with no central manager. In this method, each driver selects the route with the shortest estimated travel time among alternative routes. We employ a neural network to estimate the travel time for each route. Through computational experiments, we compare the two methods with each other and demonstrate the characteristic features of each method.
