Systematic selection of tuning parameters for efficient predictive controllers using a multiobjective evolutionary algorithm

• The objectives in MPC tuning are performance, feasibility and computational cost.
• A multiobjective problem is posed for tuning Laguerre-based MPC.
• The multiobjective problem is solved using NSGA-II.
• Experiments show that NSGA-II can obtain a suitable balance between the objectives.

In the design of predictive controllers (MPC), parameterisation of degrees of freedom by Laguerre functions, has shown to improve the controller performance and feasible region. However, an open question remains: how to select the optimal tuning parameters? Moreover, optimality will depend on the size of the feasible region of the controller, the system's closed-loop performance and the online computational cost of the algorithm. This paper develops a method for a systematic selection of tuning parameters for a parameterised predictive control algorithm. In order to do this, a multiobjective problem is posed and then solved using a multiobjective evolutionary algorithm (MOEA) given that the objectives are in conflict. Numerical simulations show that the MOEA is a useful tool to obtain a suitable balance between feasibility, performance and computational cost.

Figure optionsDownload as PowerPoint slide