Distributed MPC for urban traffic networks: A simulation-based performance analysis

摘要

The operation of urban traffic networks with distributed model predictive control (DMPC) can be more flexible than centralized strategies because DMPC allows for graceful expansion of the control infrastructure, localized reconfiguration, and tolerance to faulty operation. Yet, computational performance is less efficient than with centralized model predictive control (MPC) because of the added complexity brought about by distribution schemes. To assess the trade-off between flexibility and performance, in this paper we assess the tolerance to failure and performance of DMPC in contrast with centralized control. The problem of concern is the signal setting of green time in a representative traffic network modeled in a commercial microscopic traffic simulator. A software tool is developed for implementing and simulating the DMPC framework in tandem with the simulator. Comparisons of DMPC with MPC and a baseline feedback control strategy that does not use constrained optimization show that DMPC can achieve performance gains with respect to the baseline case and enhance tolerance to failure. Computations for DMPC are less efficient than with centralized MPC; nevertheless, the time taken by DMPC is well below the required for field use. Although the true distributed deployment of DMPC requires special hardware, its implementation in a central cluster can be made without loss of operational flexibility. Copyright (c) 2014 John Wiley & Sons, Ltd.