Comparing Dynamic Programming Based Algorithms in Traffic Signal Control System

In this paper, we mainly focus on a comparison of three types of dynamic programming based algorithms for optimal and near-optimal solutions of traffic signal control problem. The algorithms are backward dynamic programming (BDP), forward dynamic programming (FDP), and approximate dynamic programming (ADP). The traffic signal control model at isolated intersection is formulated by discrete-time Markov decision process in stochastic traffic environment. Optimal solutions by BDP and FDP algorithms are considered in traffic system for stochastic state transition and deterministic state transition, respectively. A near-optimal solution by ADP for problem control adopts a linear function approximation in order to overcome computational complexity. In simulation, these three control algorithms are compared in different traffic scenarios with performances of average traffic delay and vehicle stops.