Modeling, robust and distributed model predictive control for freeway networks

In Model Predictive Control (MPC) for traffic networks, traffic models are crucial since they are used as prediction models for determining the optimal control actions. In order to reduce the computational complexity of MPC for traffic networks, macroscopic traffic models are often used instead of microscopic traffic models. These macroscopic traffic models can be divided into homogeneous, single-class models and heterogeneous, multi-class models. In general, multi-class models are more accurate than single-class models, without increasing the computational complexity significantly. In MPC a more accurate model in general implies a better prediction of the controlled system, providing the controller more accurate information for determining the control actions. Therefore, developing and using multi-class traffic models is one way to improve the effectiveness of MPC. Apart from the above characteristics of traffic models, other factors such as uncertainties in external inputs and model parameters can also affect the accuracy of predictions. Thus another way for improving the effectiveness of MPC is to take into account the effects of these uncertainties and to develop robust MPC approaches for handling these uncertainties. Apart from improving the effectiveness of MPC, making MPC feasible for large-scale traffic networks is also important, due to the rapid increase of the computational complexity of the MPC optimization problem with the size of the controlled system. For large-scale systems, Distributed Model Predictive Control (DMPC) is often considered for making the control approach computationally feasible. Moreover, robust DMPC can be developed for ensuring both feasibility and robustness.