Controlling the NSGA-II Algorithm Convergence Toward a Fixed Pareto-optimal Solution for the Gross Domestic Product Quarterly Disaggregation
Raïmi Aboudou Essessinou and Guy A. Degla (Institute of Mathematics and Physical Sciences, University of Abomey-Calavi (UAC), Dangbo, Benin)
In this paper, we test the convergence speed of the fast elitist Non-Dominated Sorting Genetic Algorithm (NSGA-II) on the Gross Domestic Product (GDP) quarterly disaggregation problem. In fact, we perform many simulations by considering various inputs of the NSGA-II parameters with respect to the population size (pop_size), the iteration number (no_rum), the maximum number of generations (gen_max) and the mutation distribution index in the polynomial mutation (etam). It turns out that for suitably large values of the parameters pop_size, no_rum and gen_max, the NSGA-II algorithm converges to a single Pareto-optimal solution located in a fixed cuboid. Due to the elitism in the NSGA-II and our simulation results, we suspect that the mutation distribution index alone doesn't influence the time of algorithm convergence and the number of Pareto-optimal solutions. Therefore, we reach the most likely conclusion that only an appropriate choice of the parameter triplet (pop_size, no_rum, gen_max) can ensure the convergence of the algorithm toward a fixed Pareto-optimal solution for the Quadratic Multiobjective Programming of GDP quarterly disaggregation. It is worth noted that our work is an extension of previous works in the same field.
Conference: UKSim-AMSS 22nd International Conference on Computer Modelling and Simulation, UKSim2020
Published: Mar 25, 2020