Benchmark of genetic algorithms applied in the tuning of a non-linear predictive controller

This study evaluates the performance of three selection methods applied in genetic algorithms for the tuning of a nonlinear predictive controller, focusing on a closed-loop simulated thermal plant. The objective was to find the optimal weights of the weighting matrices that regulate the system response, improving stability and convergence time to a reference signal. Through an inductive-deductive methodology, simulations were implemented with different population and generation configurations, using parallel computing techniques to reduce the computational effort. The results show significant differences between the methods, highlighting the fitness-proportional selection for its lower absolute error rate and higher consistency. It is concluded that the choice of selection method directly influences the efficiency of controller tuning, suggesting its relevance in advanced control applications where precise optimization is required in non-linear systems.