A Modified NSGA-II Algorithm for Solving Simulation-based Bi-Objective System Effectiveness Optimization of UAV Swarm

System effectiveness optimization of unmanned aerial vehicles (UAV) swarm indicators based on the simulation system is essential for the UAV swarms' real-world task performance and overall cost of the system. This simulation-based effectiveness optimization problem has the features of multi-objective, mixed-variables, and decision-maker' objective preferences on effectiveness over the total cost. In this paper, a bi-objective system effectiveness model is established to optimize the swarm system's indicator. By analyzing the problem-specific features, we propose a modified NSGA-II algorithm that merges the preference of decision-maker and co-evolves the mixed-variables in the evolution process. This algorithm integrates the preference for effectiveness into the environmental selection mechanism based on the reference point and preference radius. Furthermore, a relaxation-based co-evolution approach is designed to optimize the discrete and continuous variables simultaneously. To relieve the loss of relaxation and improve the diversity of the population, a local search method based on the hypervolume metrics for the Pareto set is adopted. The comparison and ablation experiments verify the effect of the designed components of the proposed algorithm and demonstrate the algorithm's strong convergence and diversity in the preference area.