Blockchain Deep Reinforcement Learning Digital Twin Federated Learning Mobile Edge Computing (MEC) Computational Mechanics Machine Learning
Task offloading in Mobile Edge Computing (MEC) is challenging in practice than most theoretical frameworks suggest. Edge servers proceed at uneven loads, users move constantly, and wireless channels shift in ways that are hard to predict. Analytical optimization methods lean to break down under these conditions, and purely online reinforcement learning methods are time-consuming to converge because they need large exploration in live systems. Previous research has tackled pieces of this problem with digital twin-assisted learning, federated DRL (Deep Reinforcement Learning), blockchain-secured FL (Federated Learning), but integrating all four mechanisms together in a single coherent design is missing in the existing literature. This research proposes a closed-loop architecture that integrates Digital Twin (DT) simulation, Multi-Agent Proximal Policy Optimization (MAPPO), FL, and blockchain verification for MEC task offloading. The DT allows agents to evaluate decisions in a virtual environment before to act on the real system. MAPPO manages decentralized policies with a shared critic. FL combines server experiences every three iterations through reputation-weighted Federated Average (FedAvg), and a consortium blockchain filters low-quality updates while preserving a transparent trust record. In a simulated environment with 10 mobile users and 4 edge servers, the framework converges 43% faster than a non-federated PPO (Proximal Policy Optimization) baseline, gains a 14.7% higher peak reward, and carries blockchain reputation from 0.70 to a stable range of [0.920, 1.000] by iteration 50 with one random seed.
Related links
Details
Title
Unified DT–FL–DRL–Blockchain Framework for MEC Offloading
Publication Details
Software Engineering of Emerging Technologies;...Proceedings, Part I, pp.114-137
Resource Type
Conference proceeding
Conference
International Conference on Software Engineering of Emerging Technologies (SEET 2026), 2nd (Behrend, Pennsylvania, USA, 08/10/2026–08/11/2026)
Publisher
Springer Nature Switzerland; Cham
Series
Communications in Computer and Information Science