Failure Power Outage Prediction Based on Reliability Indicators: A Case Study of the Cigarette Factory Stripping Workshop

Smart manufacturing in tobacco production requires proactive power outage prediction to avoid costly disruptions. While machine learning has improved grid-level outage forecasting, workshop-specific solutions remain scarce. This study develops a reliability indicator framework for tobacco processing workshops, combining Analytic Hierarchy Process (AHP) for feature selection with optimized Relevance Vector Machine (RVM) models. AHP identifies critical metrics—branch overcurrent frequency, voltage fluctuation rate, and alarm frequency—while two enhanced RVMs (IFA-RVM and SA-PSORVM) are tested on four years of operational data. Results show SA-PSO-RVM outperforms baseline RVM and IFA-RVM, achieving 86.9% accuracy and 0.924 AUC with faster convergence. The framework bridges the gap between grid-level prediction models and industrial workshop needs, offering a practical solution for pre-fault warnings. Future research will integrate deep learning for improved nonlinear time-series analysis.