A framework for adaptive tuning of distributed model predictive controllers by Lagrange multipliers

In this work we show that some sort of altruism between controllers is require for a distributed approach to be globally optimal. This paper makes a contribution to the state-of-the-art by defining distributed MPC controllers as altruistic MPC agents and proposing an on-line tuning of the agent altruism (Lagrange multipliers). The tuning process will guarantee a minimal level of what we call satisfaction, for all MPC agents. The tuning adapts to the current conditions since it is performed in each control cycle. Further, a bargain scheme can be developed to deal with infeasibility.