A Computationally Efficient Non-cooperative Distributed Model Predictive Control Algorithm for Two Agent Systems

This paper presents a non-cooperative distributed model predictive control algorithm used in cyber-physical systems for two agent systems coupled through the inputs. Each agent computes the optimal input trajectory for its corresponding subsystem as the minimum of a local optimization problem. The input trajectory of the neighbor, which is used in the local optimization problem, is obtained based on the previous optimal control sequence and is received in a communication session. After that, the optimization problem is solved and the optimal input is sent to the process. This approach is computationally efficient because the communication between the two agents is reduced at minimum (only one session each sampling period) and the optimal input is obtained solving one optimization problem thus diminishing the overall computational time. The algorithm was implemented in Matlab and the obtained performances were compared with a cooperative distributed model predictive control strategy. Both methods were tested in simulation on a quadruple tanks process and the results recommend the non-cooperative strategy which has similar results with less computational requirements.