Avaliação do algoritmo de Luus-Jaakola para aprimoramento da sintonia de controladores PID

Abstract

To improve the tuning of the PID controllers, the Luus-Jaakola algorithm method was compared with manual tuning, random search, and grid search to improve proportional, integral, and derivative parameters. Through usual industrial systems, simulations were made with the initial manual tuning, taken as a basis for comparative character. In all systems with their standard results, the consequent evaluations of the search algorithms were performed, which resulted in certain iteration counts looking for an improved set of values for the PID controller. In all systems: all methods had similar execution time, regardless of the instructions required for each method, reaching a linear characteristic of iteration by simulation time of R² > 0.98; Luus-Jaakola consistently surpasses the efficiency of random search by providing sufficient population size and iterations for the algorithm; Luus-Jaakola managed to achieve better improvements in a more complex system; the network search algorithm can be more efficient than Luus-Jaakola and the random search if the initial search space and its size is adequate.