Incorporating Kinematic Wave Theory Into a Deep Learning Method for High-Resolution Traffic Speed Estimation

We propose a kinematic wave-based Deep Convolutional Neural Network (Deep CNN) to estimate high-resolution traffic speed fields from sparse probe vehicle trajectories. introduce two key approaches that allow us incorporate wave theory principles improve the robustness of existing learning-based estimation methods. First, we an anisotropic kernel for CNN. The explicitly accounts space-time correlations in macroscopic and effectively reduces number trainable parameters CNN model. Second, use simulated data training Using targeted provides implicit way impose desirable physical features on learning In experiments, highlight benefits using kernels evaluate transferability trained model real-world Next Generation Simulation (NGSIM) German Highway Drone (HighD) datasets. results demonstrate significantly reduce complexity over-fitting, correctness estimated fields. find scales linearly with problem size compared quadratic scaling isotropic kernels. Furthermore, evaluation datasets shows acceptable performance, which establishes simulation-based is viable surrogate data. Finally, comparison standard techniques superior accuracy proposed method.