Time-window-constrained computation offloading via cost-efficient multi-UAV scheduling in space-air-ground integrated networks

Abstract

In air-ground collaborative sensing scenarios such as periodic environmental monitoring, emergency rescue, and military reconnaissance in remote areas, computation tasks are often subject to strict time window constraints, urgently necessitating efficient computing mechanisms to guarantee their timeliness. We investigate the joint optimization problem of computation offloading and unmanned aerial vehicle (UAV) scheduling under time-window constraints within a collaborative computing architecture involving UAVs and high-altitude platform (HAP). To reduce the number of dispatched UAVs and their energy consumption, a mixed-integer nonlinear programming problem is formulated aiming at minimizing the execution cost of computation tasks. Addressing the core challenge posed by the non-convex strong coupling between discrete and continuous variables, we propose a novel bi-level collaborative optimization strategy (BCOS). The outer layer employs the covariance matrix adaptation evolution strategy (CMA-ES) to efficiently search for optimal task offloading ratios within the continuous domain, while the inner layer, given the offloading strategy, precisely solves the discrete UAV route planning subproblem based on the branch-and-price algorithm. Through the iterative interaction between the inner and outer layer algorithms, the joint optimization of discrete and continuous variables is achieved upon convergence. Simulation results demonstrate that BCOS can effectively coordinate the optimization of computational resources and UAVs scheduling strategies, significantly reducing the number of dispatched UAVs and energy consumption, thereby validating its effectiveness and superiority in complex task execution scenarios.