Modelling of a traffic cell based on a recurrent-neural network
Marcos Angel Gonzalez-Olvera, Yu Qing Tang, Luis A. Álvarez-Icaza · International Journal of Modelling Identification and Control · 2011
In this paper, a continuous-time recurrent neural network for modelling a vehicle density-flow relation in a section of a highway is presented. The global objective is to make the network emulate dynamically this individual section for simulation purposes. The training algorithm is motivated from previous works in adaptive observers and uses only output measurements and the knowledge of the excitation input signal to generate the entire dynamics of the network. Training is based on the generation of estimates of an ideal network and jointly identifying its parameters. The stability and convergence of the training algorithm are established based on the Lyapunov stability theory. Model validation through numerical simulation with real data is included.