A Deep Neural Network for Traffic Flow Simulation

Macroscopic road traffic simulation is based on partial differential equations for vehicle conservation and flow. To solve the road traffic problem, we propose to solve partial differential equations using deep learning and a physically informed neural network approach. The method is then applied on a road. The results show that after solving several points that can be randomly chosen on the space and time (x,t), the result obtained after training is a neural function meshless dependent function of (x,t). In this case, the precision of the estimation is independent of any mesh, something we often encounter in classical approaches to discretization by finite differences or finite volumes.