Deep Reinforcement Learning-based Predictive Maintenance Task Offloading and Resource Allocation

Abstract

In recent years, the growth of Industrial Internet of Things has enabled automated data collection and analysis, leading to the re-emerging of predictive maintenance. However, IIoT devices have insufficient computing power for computationally intensive maintenance tasks. Consequently, Mobile Edge Computing (MEC) has been introduced to relieve the computation burden. This paper focuses on predictive maintenance task offloading and resource allocation optimization based on Deep Deterministic Policy Gradient (DDPG). An Artificial Bee Colony algorithm (ABC) is further incorporated to speed up the learning convergence. The proposed algorithm minimizes the overall maintenance cost in generating the optimal task offloading and resource allocation solution. Experimental results show that this scheme outperforms the baseline scheme in reducing the total system maintenance cost.