A dynamic network loading process with explicit delay modelling

In this paper, a neural network (NN) approximator, integrated to a dynamic network loading (DNL) process, is utilized to model delays and to solve the DNL problem at an unsignalized highway node. First, a dynamic node model (DNM) is set out to compute the time-varying traffic flows conflicting at the node. The presented DNM has two components: a link model set with a linear travel time function and an algorithm written with a set of node rules considering the constraints of conservation, flow splitting rates and non-negativity. Each of the selected NN methods, feed-forward back-propagation NN, radial basis function NN, and generalized regression NN, are utilized one by one in the NN approximator that is integrated with the proposed DNM, and, hence, three DNL processes are simulated. Delays forming as a result of capacity constraint and flow conflicting at the node are calculated with selected NN configurations after calibrating the NN component with conical delay function formulation. The results of the model structure, run solely with the conventional delay function, are then compared to evaluate the performance of the models supported with NNs relatively.