TITLE:
Transformer with Sparse Mixture of Experts for Time-Series Data Prediction in Industrial IoT Systems
AUTHORS:
Feng Shi, Bolin Li, Weidong Zhang
KEYWORDS:
Transformer Model, Industrial Internet of Things, Multivariate Time-Series Prediction, Self-Attention Mechanism, Mixture of Experts
JOURNAL NAME:
Engineering,
Vol.17 No.3,
March
26,
2025
ABSTRACT: With the development of the Industrial Internet of Things (IIoT) and cloud computing technologies, intelligent sensors in the field that can generate large volumes of time-series data continuously have emerged. Due to the lack of equipment and network impacts, highly distributed industrial applications cannot capture and transfer all production data to a distant cloud server in real time. Consequently, a portion of critical production data is lost, which poses the significant challenge of the timely replenishment of missing data. Employing deep learning in the cloud center for data trend prediction based on relevant data can solve this problem. The objective of this study was to develop a time-series prediction model that combines a Transformer model with a sparse Mixture of Experts (MoE). The model is designed specifically for an IIoT system that is used in oil-well operations. The proposed TransMoE prediction model combines the advantages of the MoE and the Transformer model. The MoE can effectively handle multiple subtasks while the Transformer algorithm can reflect the long-range dependency of the input data series. The proposed model was used to predict oil-well yields, and the predicted outcomes were compared with those obtained using a CNN-GRU and CNN-LSTM models, as well as the actual recorded data. The experimental results indicated that the proposed TransMoE model can significantly increase the efficiency and accuracy of oil well production sequence data prediction, with an average relative error of 6.26%, which can satisfy the requirements of enterprise data usage.