A hybrid quantum-classical architecture for combinatorial decision optimization in networked systems

Abstract

Combinatorial decision optimization problems arise widely in modern networked systems, where limited communication, computing, and service resources must be efficiently allocated under complex operational constraints. Representative examples include supply-demand matching in data markets, topology control in self-organizing Unmanned Aerial Vehicle (UAV) swarms, and microservice scheduling across the cloud-edge continuum. These problems are typically NP-hard, and as system scale increases or operating conditions evolve rapidly, traditional Mixed-Integer Linear Programming (MILP) formulations often become difficult to solve within real-time decision windows. As a unified binary optimization framework, Quadratic Unconstrained Binary Optimization (QUBO) provides a common way to map diverse combinatorial problems to quantum annealing and quantum-inspired solvers with the potential for significant computational speed advantages. However, the practical use of QUBO in real networked systems still faces three major barriers. First, QUBO modeling remains manual, expert-dependent, and error-prone. Second, standard QUBO formulations are inherently static and therefore not well suited to time-varying environments. Third, the binary representation of QUBO does not naturally align with the continuous resource allocation requirements of real systems. To address these limitations, this thesis develops a hybrid quantum-classical optimization methodology for networked systems. It first formulates and validates domain-specific QUBO models for representative applications. Then it generalizes these efforts through two-stage hybrid frameworks that combine offline combinatorial optimization with lightweight online decision-making for dynamic UAV topology control and robust microservice scheduling. Finally, it investigates large language model driven automation of the MILP-to-QUBO pipeline and integrates Benders decomposition to improve scalability for larger problem instances. Overall, this thesis shows that QUBO can serve not only as a problem-specific solution form, but also as a transferable modeling layer that connects heterogeneous network optimization tasks with near-term quantum hardware, thereby providing a practical pathway toward quantum-enhanced decision-making.