Data-based distributed model predictive control for large-scale systems

This paper investigates a data-based distributed model predictive control (DMPC) method for large-scale systems composed of a number of isolated subsystems. Under the circumstances that the system dynamics knowledge is unknown, the proposed method includes the learning phase of the system model and the design phase of the model-based controller in serial. First, a practical learning algorithm is designed utilizing the observed input–output data of the controlled plant for model learning. which consists of the projection identification procedure and neural networks (NNs) estimation procedure. Thus, the linear model of the system and the unmodeled dynamics are obtained. Based on the learned system model, the DMPC method is developed for constrained large-scale systems. The controller is derived by solving the optimization problem of isolated sub- systems, which minimizes the local cost function of each subsystem. Moreover, sufficient conditions are established for the stability of the overall system. The recursive feasibility and convergence are guaranteed by rigorous derivation. The effectiveness of the DMPC approach proposed in this paper is demonstrated by applying it to the vehicle platoon system.