Statistical convergence analysis of ACO - NM for PID controller tuning

Optimal controller and anti-windup tuning can be identified to a hard optimization problem and be solved by metaheuristics. Since metaheuristics' performance is based on the balance between the diversification and intensification processes obtained by adjusting the method parameters, it is important to set it adequately to provide a high quality solution. A statistical Ant Colony Optimization (ACO) analysis is proposed to establish the quality of the solution reached with regard to the number of ants and the number of objective function evaluations. Sensitivity curves to the number of ants and number of function evaluations for two different discretization search space are presented. For a lower number of function evaluations for ACO, a better starting point for the Nelder-Mead (NM) local search has been deteimined. The final system response is comparable to the previous ACO-NM algorithm for almost two times less evaluations of the objective function.