Advanced Genetic Algorithm-Based Network Optimization for Mission Assurance

Abstract Optimizing network configurations is essential for improving the performance of mission-critical systems. This paper presents a genetic algorithm (GA) framework that integrates a constraint-preserving, mutation-only formulation with a scalable CPU-GPU execution strategy to support efficient exploration of hierarchical network designs. The framework maintains structural feasibility during optimization and enables parallel evaluation through a GPU architecture in which independent thread blocks evolve subpopulations concurrently. A key feature of the approach is Dynamic Bracket Selection (DBS), a structured probabilistic selection mechanism inspired by tournament brackets. DBS extends existing stochastic selection schemes by introducing multi-round advancement with tunable selective pressure, allowing diversity to be maintained even in mutation-only search environments. This provides a practical alternative to conventional elitist and tournament strategies, particularly when operating under strict feasibility constraints. The framework is evaluated on a simplified electrical-grid model designed to isolate optimization behavior rather than replicate full physical power-flow dynamics. Comparative experiments examine CPU and GPU performance, block-level parallelism, and the influence of different selection strategies. Results show that GPU parallelization accelerates early convergence and that DBS supports broader search exploration while producing competitive fitness outcomes. Together, these elements demonstrate how combining structured probabilistic selection with GPU-enabled parallelism can improve the efficiency and adaptability of evolutionary search in constrained networked systems, offering a foundation for design-stage analysis in complex engineering applications.