Resource Scheduling for UAV-Assisted Failure-Prone MEC in Industrial Internet

• Published in: Drones (Basel) Vol. 7; no. 4; p. 259
• Main Authors: Li, Xuehua, Fang, Yu, Pan, Chunyu, Cai, Yuanxin, Zhou, Mingyu
• Format: Journal Article
• Language: English
• Published: Basel MDPI AG 01.04.2023
• Subjects: Algorithms; Communication; Communications networks; Data integrity; Deep learning; deep reinforcement learning; Design optimization; Drone aircraft; Edge computing; Energy consumption; Failure; Industrial Internet; Integer programming; Internet; Linear programming; Mathematical models; Mathematical optimization; Methods; Mobile computing; mobile edge computing; multi-UAVs; Network function virtualization; Network latency; Network reliability; Optimization; Real time; Reinforcement learning (Machine learning); Resource allocation; Resource scheduling; Scheduling (Management); Search methods; Servers; Software; Task scheduling; virtual network function; Virtual networks; Virtualization; China
• ISSN: 2504-446X; 2504-446X
• DOI: 10.3390/drones7040259

This paper focuses on reducing execution delays of dynamic computing tasks in UAV-assisted fault-prone mobile edge computing (FP-MEC) systems, which combine mobile edge computing (MEC) and network function virtualization (NFV) technologies. FP-MEC is suited to meet Industrial Internet (IIN) requirements such as data privacy, low latency, and low-cost industrial scalability in specific scenarios. However, the reliability of virtual network functions (VNFs) deployed on UAVs could impact system performance. Thus, this paper proposes the dynamic task scheduling optimization algorithm (DTSOA) based on deep reinforcement learning (DRL) for resource allocation design. The formulated execution delay optimization problem is described as an integer linear programming problem and it is an NP-hard problem. To overcome the intractable problem, this paper discretizes it into a series of single-time slot optimization problems. Furthermore, the experimental rigor is improved by constructing a real-time server state update system to calculate the real-time server load situation and crash probability. Theoretical analysis and experiments show that the DTSOA has better application prospects than Q-learning and the recent search method (RSM), and it is closer to the traversal search method (TSM).