Enhancing Adaptability and Efficiency of Task Offloading by Broad Learning in Industrial IoT

In the Multi-access Edge Computing (MEC)-based Industrial Internet of Things (IIoT), a key challenge is to make an efficient task-offloading decision. Machine learning methods have emerged as popular solutions to address this issue. However, in IIoT, it is common for the feature distribution of data to change significantly over time, i.e., data drift, and existing machine learning-based schemes struggle to frequent data drift, failing to maintain consistent high accuracy of task-offloading decisions. This struggle arises because they require extended retraining or extensive model adjustments, which involve significant delays and increased computational overhead due to the complex network structure. In this paper, we propose a <bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">B</b>road learning-based task <bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">OFF</b>loading scheme (BOFF). In BOFF, a data drift detection method based on statistical features and a sliding window is established to determine the occurrence of data drift in the system, while utilizing the Gini coefficient to enhance feature extraction and improve accuracy of task-offloading decision model under data drift. When data drift is detected, BOFF leverages its fast training and redeployment capabilities based on feature-enhanced broad learning to update the task offloading model and maintain accuracy. In the absence of significant data drift, minor changes in data distribution are addressed through incremental updates to slow the decline in model accuracy. Numerical results demonstrate that BOFF significantly improves the adaptability of data drift, ensuring high accuracy and efficiency of task offloading in dynamic IIoT environments.