A comparative study on numerical optimization techniques for Model Predictive Variable Speed Limit control

Freeway congestion is highly recognized as a worldwide traffic problem since traffic demand has grown steadily over the past decades. Active traffic and demand management (ATDM) methods, including Ramp Metering (RM), Variable Speed Limit (VSL), and Route Guidance (RG), can help delay or even avoid congestion. Currently, those measures have become predictive and integrated based on real-time data collection and facility coordination, e.g. Model Predictive Control (MPC)-based measures. In this generic approach, a second-order macroscopic traffic flow model is used for traffic state prediction and the control problem is formulated as a nonlinear optimization problem; whereas, successful field implementation requires a computationally simple and accurate optimization technique to make it feasible in practice. In previous applications of MPC, common optimization techniques are the decision tree and sequential quadratic programming (SQP). They both perform well, but limited research has been conducted to assess and compare their effectiveness and efficiency. To address this research gap, this study tested their performance by applying an MPC-based VSL control. Using geometric and traffic data from an authentic freeway corridor, this paper discussed the speed limit sequences, measures of effectiveness (MOEs), and computation time from micro-simulation tests. The results of this comparative study can guide future filed implementation as a reference.