Multi‐Mode Data Prediction‐Based Dynamic Bandwidth Allocation for IoT‐Empowered Electric Equipment Monitoring

ABSTRACT With the gradual increase in the demand for efficient monitoring of electrical equipment, the volume of multi‐mode data, such as images and voiceprints, is also growing. This situation imposes new challenges on the allocation of limited bandwidth resources. Traditional allocation methods suffer from issues such as low bandwidth utilisation and mismatch between bandwidth resources and multi‐mode data transmission demands. To address these problems, a dynamic bandwidth allocation method based on multi‐mode data prediction for application in internet of things (IoT)‐enabled electrical equipment monitoring is proposed in this paper. Firstly, a multi‐mode data transmission architecture for IoT‐enabled electrical equipment monitoring is designed, which includes models for multi‐mode data collection, compression, transmission, decoding, and fault identification. Secondly, a multi‐mode data transmission demand prediction method based on knowledge collaborative long short‐term memory (KC‐LSTM) is proposed, considering the intrinsic relationships and complementarity among multi‐mode data streams to achieve accurate prediction of multi‐mode data stream transmission demands. On this basis, a dynamic bandwidth allocation method based on multi‐mode data transmission demand‐aware deep actor critic (DAC) is proposed, which dynamically allocates transmission bandwidth according to the prediction results of multi‐mode data transmission demands. Meanwhile, by constructing a multi‐precision experience replay pool, the convergence performance of the algorithm in dynamic and challenging environments is improved. Simulation results demonstrate that the proposed algorithm achieves optimal multi‐mode data stream transmission efficiency and the highest fault identification accuracy. Compared to the three benchmark algorithms, the proposed algorithm achieves 12.35%, 17.91%, and 31.84% improvements in successfully decoded data volume and 49.04%, 56.38% and 26.53% enhancements in load balancing performance, respectively.