A. Jamshidnejad, G. Gomes, A.M. Bayen, and B. De Schutter, "Integrated offline and online predictive control system within a base-parallel architecture," Tech. rep. 18-025, Delft Center for Systems and Control, Delft University of Technology, Delft, The Netherlands, 2018. Submitted for publication. See also https://arxiv.org/abs/1907.05464v1.
Optimization-based controllers minimize a specific performance index within a finite or infinite prediction window, and find the corresponding optimal control input. An estimator helps the controller to determine the future states of the controlled system and the external inputs such as disturbances, etc. Online application of optimization-based controllers is still a challenge, especially since the time required for investigating the search space by the optimization solver usually exceeds the requirements of a real-time procedure. In this paper, we propose a novel integrated control architecture that benefits from the advantages of both offline and online controllers within a predictive base-parallel structure. The base block includes efficient parameterized controllers, which have been optimized offline w.r.t. their parameters within an update window, which is larger than the prediction window, and direct controllers, which have been trained offline to produce a sequence of control inputs within the prediction window. These controllers provide good starting points for the optimization-based controllers that are run in parallel in real time. We discuss different options for designing the proposed base-parallel control architecture. Finally, we implement the proposed architecture for a highway that is controlled by ramp metering, and compare our results with the results of previous efficient controllers, such as ALINEA and an artificial neural network-based controller that has been trained by deep learning.