Suboptimal search strategies with bounded computational complexity to solve long-horizon direct model predictive control problems

Search algorithms that reduce the time to solve the direct model predictive control (MPC) problem are proposed in this paper. By allowing for suboptimal solutions, the computational complexity of the underlying optimization problem can be significantly reduced, albeit by sacrificing (to a certain degree) optimality. Two approaches are presented and discussed. The first approach requires quadratic time, making it a very efficient candidate for solving the examined problem. Thanks to the second approach, a preset upper limit on the operations performed in real time is not exceeded, thus guaranteeing realtime termination in all runs. To highlight the effectiveness of the introduced strategies, a variable speed drive system with a three-level voltage source inverter is used as an illustrative example.