Traffic congestion on freeways is a critical problem due to its negative impact on the environment and many other important consequences like delays, waste of fuel, a higher accident risk probability, etc. Freeways were originally conceived to provide virtually unlimited mobility to road users. However, the continuous increase in car ownership and demand has led to a steady increase (in space and time) of recurrent and non-recurrent freeway congestion, particularly within and around metropolitan areas. The construction of new freeways is not always a viable solution that can be implemented in the short term due to technical, political, legal, or economic reasons. Therefore, in the last decades, a lot of research has been focused on making a better use of the available traffic infrastructure. It has been demonstrated in the literature that dynamic traffic control is an excellent solution to decrease congestion. In general, dynamic traffic control uses measurements of the traffic conditions over time and computes dynamic control signals to influence the behavior of the drivers and to generate a response in such a way that the performance of the network is improved, by reducing delays, emissions, fuel consumption, etc. Variable Speed Limits (VSL), ramp metering, and reversible lanes are some of the most often used examples of dynamic freeway traffic control measures. These measures have been already successfully implemented in Germany, Spain, The Netherlands, US, Australia, France and other countries.

Nowadays, most of the dynamic traffic control systems implemented operate according to linear, local or heuristic control algorithms. However, the use of appropriate non-local and multivariable techniques can improve considerably the reduction in the total time spent by the drivers and other traffic performance indices. Among the available options described in the literature, the methods based on the use of advanced control techniques like Model Predictive Control (MPC), which minimizes a cost function like the total time spent by the drivers, have shown to substantially improve the performance of the controlled traffic system in various simulation studies. The main problem of MPC is that the computation time quickly increases with the size of the network, making it difficult to apply centralized MPC for large networks. For this and other reasons, completely centralized control of large networks is viewed by most practitioners as impractical and unrealistic. In order to overcome this practical problem, easy-to-implement control algorithms have been designed for ramp metering and reversible lanes. However, an easy-to-implement VSL control algorithm which approximates the performance of an MPC controller has to be necessarily a bit more complex.

The main objective of this project is the design and testing of a control algorithm for VSL that approaches the behavior of an optimal controller and can, at the same time, be applied in practice to large traffic networks. When designing this controller, it has to be taken into account in general that a linear or logic-based controller for VSL, which can performs properly for one particular kind of congestion, is not going to approach the MPC behavior for other kinds of congestion. Therefore, this proposal suggests the use of two control levels. In the upper level, a scheduling controller detects online the main kinds of congestion (recurrent congestions, shock waves and unexpected capacity reductions) and, in the lower level, a practically implementable controller for each kind of congestion is used. The proposed algorithm will be tested using two freeways: SE-30 in Seville, Spain and A12 in South Holland, The Netherlands.