A3CMulti-Edge: Multi-Agent Cross-Edge-Cloud Collaborative Task Scheduling Policy for Underground Coal Mine Intelligent Monitoring

Wei Chen, Zike Ma, Tingting Xu, Xin Li, Jueting Liu, Zehua Wang

Abstract: Intelligent monitoring in underground coal mines is challenged by limited computational resources, dynamic task workloads, and unstable communication networks. Traditional centralized architectures struggle to meet the demands of low latency and high throughput under such constraints. To address these challenges, we propose A3CMulti-Edge, a multi-agent cross-edge-cloud scheduling policy designed for IIoT environments. A3CMulti-Edge (hereafter referred to as A3C-Edge) integrates edge and cloud computing in a cloud–edge collaborative architecture and leverages Markov game modeling with the A3C algorithm to learn adaptive task scheduling policies. To enhance performance, we incorporate entropy regularization for improved exploration and an action masking mechanism to ensure valid task-to-node assignments, improving both resource utilization and system robustness. We evaluate the proposed policy on a SimPy-based simulation platform using the cluster-trace-v2018 dataset. Experimental results show that A3C-Edge achieves a throughput of 87%, outperforming baseline methods such as EGDRL, MEETS-RL, MBE-RNN, and MOMFO in terms of throughput, resource allocation efficiency, and load balancing. Furthermore, A3C-Edge demonstrates greater stability under conditions of load fluctuation and network uncertainty. These findings suggest that A3C-Edge provides an effective and resilient scheduling solution for underground IIoT monitoring systems.

External IDs:dblp:conf/icic/ChenMXLLW25