Correlated Information Scheduling in Industrial Internet of Things Based on Multi-Heterogeneous-Agent-Reinforcement-Learning
Qiuyang Zhang, Ying Wang
Abstract
The industrial Internet of Things (IIoT) has led to the emergence of various information-based industrial applications. Due to the uninterrupted and complex nature of production processes in industrial systems, these applications often run continuously and rely on information from multiple sensors. As a result, a single sensor can support multiple applications simultaneously, leading to complex correlations in the system. To address this challenge, we introduce the concept of the age of correlated information (AoCI) and formulate the scheduling problem as a Markov game problem to optimize the information freshness of industrial applications. To solve the problem, we propose a multi-heterogeneous-agent-reinforcement-learning (MHARL) scheme, which uses neural networks with different structures to represent agents participating in the game. Our numerical results demonstrate that the proposed MHARL scheme outperforms typical baselines, such as Qmix and VDN, in terms of AoCI and energy efficiency.