A real-coded genetic algorithm involving a hybrid crossover method for power plant control system design

This paper introduces a new hybrid crossover method for a real-coded genetic algorithm and its application to control system design of a power plant. Determining gains for controllers by using a genetic algorithm method usually involves multiple training stages. This method is not necessarily optimal. This paper applies a hybrid crossover method in a real-coded genetic algorithm to simultaneously find gains of three PI control loops and six other coupled gains in a boiler-turbine control system. The real-coded genetic algorithm with the hybrid crossover method has a better convergence rate when applied to this problem, as compared to other methods. A better convergence rate reduces execution time and is particularly relevant to problems having significant simulation times. A comparison between hybrid crossover and convex crossover in a real-coded genetic algorithm together with multi point crossover using a binary coded genetic algorithm has also been made