Research on Intelligent Perception and Adaptive Operation and Maintenance of Electrical Systems Empowered by Industrial Internet of Things Based on Transformer

Aiming at prominent issues in traditional industrial electrical system operation and maintenance operation and maintenance(O&M) such as imprecise state perception, difficulty in fault prediction, and low O&M efficiency, this paper proposes a new method for intelligent electrical system state perception and adaptive O&M by integrating the Industrial Internet of Things Industrial Internet of Things(IIoT) and the Transformer model. Based on the three-layer architecture of IIoT, the core of this method is to use the Transformer model to process multi-source heterogeneous sensor data. Through its powerful capabilities in sequence feature extraction and long-range dependency capturing, it achieves accurate dynamic perception of electrical equipment operating status and early fault warning, thereby driving adaptive O&M decisions. Experimental verification on public datasets (such as the University of California, Irvine(UCI) Hydraulic System dataset and the IEEE PHM Data Challenge motor fault dataset) shows that the Transformer-based method proposed in this paper achieves an $\mathbf{F 1}$-score (macro average) of $\mathbf{9 5. 6 \%}$ in state recognition tasks, an improvement of about 5.4 percentage points compared to Long Short-Term Memory(LSTM). In electrical equipment fault prediction tasks, the $\mathbf{F 1}$-score reaches $\mathbf{9 1. 1 \%}$, significantly outperforming baseline comparisons. These data indicate that the proposed method can greatly improve the accuracy of fault prediction and effectively reduce unplanned equipment downtime (e.g., downtime reduced from $\mathbf{4 8}$ hours per month to $\mathbf{1 0}$ hours), providing a strong technical solution for the intelligent and efficient transformation of electrical system O&M.