Weight optimisation for iterative distributed model predictive control applied to power networks

Reference:
P. Mc Namara, R.R. Negenborn, B. De Schutter, and G. Lightbody, "Weight optimisation for iterative distributed model predictive control applied to power networks," Engineering Applications of Artificial Intelligence, vol. 26, no. 1, pp. 532-543, Jan. 2013.

Abstract:
This paper presents a weight tuning technique for iterative distributed Model Predictive Control (MPC). Particle Swarm Optimisation (PSO) is used to optimise both the weights associated with disturbance rejection and the weights associated with achieving consensus between control agents (while this paper focuses on disturbance rejection, the same techniques could also be used for set-point tracking based weight optimisation). Unlike centralised MPC, where tuning focuses solely on disturbance rejection performance, iterative distributed MPC practitioners must concern themselves with the trade off between disturbance rejection and the overall communication overhead when tuning weights. This is particularly the case in large scale systems, such as power networks, where typically there will be a large communication overhead associated with control. This paper examines the effects of weight optimisation on both the disturbance rejection and the communication overhead. Two PSO fitness functions are employed; the first function evaluates fitness based solely on disturbance rejection ability, and the second is based on achieving a trade off between good disturbance rejection ability and the maximum number of distributed MPC iterations per control step. Simulation experiments illustrate the potential of the proposed approach for weight tuning in two different power system scenarios.