Online state and multidimensional parameter estimation for a macroscopic model of a traffic junction

An online multidimensional self-estimation algorithm is developed to jointly estimate the parameters of a macroscopic model describing the traffic dynamics in a signalized junction under different traffic conditions, together with the state variables characterising traffic flow. The proposed novel methodology is based on the Expectation-Maximization algorithm and multidimensional Robbins-Monro stochastic approximation. The algorithm is validated on the geometry of a signalized 3-arm junction within the traffic network of Malta resulting in a mean percentage error of -0.965% on the parameter estimates. This is aimed to form part of an adaptive control loop for traffic light systems that is able to autonomously adjust to changing traffic conditions.