Solving a maximum flow problem using backpropagation

One of the most commonly used methods to quantify traffic in terms of flow and speed is generally known as the Greenberg model. The authors propose to relate some of the parameters that are used to compute the Greenberg equations by mapping them by means of a neural network. It is noted that many different aspects of the relationship between the traffic flow and the traffic density as expressed in the Greenberg model can be mapped by means of the neural net. In the case considered, the authors choose to relate the maximum traffic flow (vehicles/minute) for different traffic densities. As input for the net they used the traffic density (number of vehicles/unit length) and the space mean speed; the output will be the maximum possible traffic flow. The simulation discussed indicates that the relations between different traffic parameters can be adequately learned by a neural network