AI-Driven Edge–Cloud Orchestration of Digital Twins over 5G Networks for Latency-Critical Emergency Response
Main Article Content
Abstract
Background: Digital Twins (DTs) have become a central instrument for situational awareness in smart cities, and emergencies are precisely the conditions under which a twin's value is greatest and its computational cost hardest to meet.
Problem and Research Gap: Existing systems resolve DT placement statically, pinning models either to resource-limited edge nodes or to distant clouds irrespective of the fluctuating network and computational conditions that characterise a disaster. Recent studies of DT offloading optimise a single objective, treat 5G Quality of Service (QoS) as an exogenous constant rather than an observable control signal, and disregard the differentiated urgency of emergency tasks.
Proposed Method: This paper proposes EDGE-DTO, an AI-driven orchestration framework that continuously decides whether each DT computation should execute at the edge, in the cloud, or in a split hybrid configuration. It couples a multi-agent proximal policy optimisation controller with an NSGA-II multi-objective optimiser supplying Pareto-efficient warm-start policies, and consumes live 5G QoS indicators—5QI class, round-trip delay, achievable uplink rate, and slice reliability—together with edge utilisation, cloud queue depth, and an incident priority weight.
Results: An agent-based simulation of citizens, responders, vehicles, UAVs, and infrastructure agents evaluates the framework across earthquake, flood, wildfire, and industrial-accident scenarios. Mean end-to-end latency falls by 33–54 percent relative to static and threshold-based baselines and by 22 percent relative to a single-agent deep reinforcement learning baseline, with deadline adherence for high-priority twins reaching 96.4 percent.
Contribution: The work reframes DT placement as a continuous, QoS-informed control problem, exposes the latency–energy–QoS trade-off as an operator-selectable Pareto front, and specifies a campus-scale pilot for empirical validation.