Distributed model predictive control - IEEE Control Systems Magazine

In model predictive control (MPC), also called receding horizon control, the control input is obtained by solving a discrete-time optimal control problem over a given horizon, producing an optimal open-loop control input sequence. The first control in that sequence is applied. At the next sampling instant, a new optimal control problem is formulated and solved based on the new measurements. The theory of MPC is well developed; nearly all aspects, such as stability, nonlinearity, and robustness, have been discussed in the literature (see, e.g., [1]-[4]). MPC is very popular in the process control industry because the actual control objectives and operating constraints can be represented explicitly in the optimization problem that is solved at each control instant. Many successful MPC applications have been reported in the last two decades [2], [4]. Typically, MPC is implemented in a centralized fashion. The complete system is modeled, and all the control inputs are computed in one optimization problem. In large-scale applications, such as power systems, water distribution systems, traffic systems, manufacturing systems, and economic systems, it is useful (sometimes necessary) to have distributed or decentralized control schemes, where local control inputs are computed using local measurements and reduced-order models of the local dynamics [5], [6]. The goal of the research described here is to realize the attractive features of MPC (meaningful objective functions and constraints) in a decentralized implementation. Previous work on distributed MPC is reported in [7]-[14]. In some applications, multiple low-level controllers are simply implemented using MPC, just as one might use proportional-integral-derivative (PID) controllers to close local feedback loops [13]. For water distribution systems, full-scale centralized MPC computations have been decomposed for decentralized computation, using standard coordi-