Distributed and Data-Driven Optimization Frameworks for Logistics-Oriented Decision Support Under Partial and Asynchronous Information

This paper introduces D3O-GT, a distributed optimization framework designed to operate under partial, heterogeneous, and delayed information—conditions commonly encountered in large-scale logistics and networked decision support systems. The proposed approach integrates gradient tracking with delay-aware updates to address the steady-state bias and instability that often affect classical distributed gradient methods. We formulate a consensus optimization model that captures decentralized decision variables while preserving global optimality, and we develop an algorithmic structure that balances convergence accuracy, communication efficiency, and robustness to asynchronous updates. Extensive numerical experiments demonstrate that D3O-GT achieves machine precision convergence in synchronous settings and remains stable under bounded communication delays, converging to a small neighborhood of the optimum. In contrast, conventional distributed gradient descent exhibits significant residual error under the same conditions. Scalability analyses further indicate that the proposed method maintains favorable iteration complexity as the number of agents increases. These results position D3O-GT as a practical and scalable solution for distributed decision-making environments, with direct relevance to logistics-oriented applications such as resource allocation, coordination of networked services, and real-time operational planning.