BLSO: Broad Learning System-based Scheme for Adaptive Task Offloading in Industrial IoT

In the Industrial Internet of Things (IIoT) enabled by multi-access edge computing (MEC), numerous heterogeneous devices are capable to offload computing-heavy tasks to MEC servers, thereby reducing the network congestion and accelerating task processing. Since the tasks are constrained by deadlines, the main goal of making a task offloading decision is to minimize the total time consumption of the system. Machine learning (ML) have been widely employed to address this problem. However, the popular schemes like deep learning and reinforcement learning require extensive training time, impacting the promptness of task offloading decisions. Additionally, when new devices enter the system, these schemes become unsuitable and necessitate retraining. To solve the problem, this paper proposes a broad learning system-based task offloading scheme (BLSO) to make task offloading decisions. Compared with previous works, BLSO significantly reduces training time while preserving the accuracy of task offloading decision. Moreover, when the environment changes frequently such as upon the addition of new devices joining in or the data distribution of tasks has been changed, BLSO can be readily updated, maintaining adaptiveness to environmental changes. Experimental results demonstrate that BLSO considerably enhances the efficiency of task offloading decision-making with high accuracy.