Traffic congestion causes unnecessary delay, pollution and increased fuel consumption. In this thesis we address this problem by proposing new algorithmic techniques to reduce traffic congestion and we contribute to the development of a new Intelligent Transportation System. We present a method to determine speed limits, in which we combine a traffic flow model with reinforcement learning techniques. A traffic flow optimization problem is formulated as a Markov Decision Process, and subsequently solved using Q-learning enhanced with value function approximation. This results in a single-agent and multi-agent approach to assign speed limits to highway sections. A difference between our work and existing approaches is that we also take traffic predictions into account. The performance of our method is evaluated in macroscopic simulations, in which we show that it is able to significantly reduce congestion under high traffic demands. A case study has been performed to evaluate the effectiveness of our method in microscopic simulations. The case study serves as a proof of concept and shows that our method performs well on a real scenario.