RELIABILITY-AWARE EDGE INTELLIGENCE WITH D3QN FOR FAULT-TOLERANT COMPUTATION IN HEALTHCARE INTERNET OF THINGS

Abstract

Healthcare Internet of Things (H-IoT) applications require timely and reliable computation for continuous patient monitoring and emergency healthcare services. Although Mobile Edge Computing (MEC) significantly reduces computation latency by offloading tasks from resource-constrained healthcare devices, existing offloading approaches primarily optimize latency and energy consumption while overlooking the reliability of edge servers. Consequently, tasks may be assigned to overloaded or failure-prone MEC servers, resulting in increased processing delays, higher task loss, and degraded service reliability. To address this issue, this paper proposes a Reliability-aware Dueling Double Deep Q Network (RA-D3QN) for fault-tolerant computation offloading in H-IoT systems. The proposed framework continuously evaluates the reliability of MEC servers based on their computational workload, queue occupancy, communication quality, and historical failure behavior, and incorporates these reliability indicators into the deep reinforcement learning process to enable adaptive and reliable offloading decisions. Extensive simulations are conducted under varying numbers of H-IoT devices and MEC servers and compared with Local Execution, Random Offloading, Greedy Latency-Based Offloading, DQN, and Double DQN. The simulation results demonstrate that the proposed RA-D3QN reduces the average latency by up to 12.6%, energy consumption by 10.9%, and task loss rate by 22.6% compared with Double DQN, while improving the average system reliability by approximately 9.5%. These results demonstrate the effectiveness of the proposed framework in providing efficient and reliable computation offloading for fault-tolerant Healthcare IoT systems