Abstract
<title>Abstract</title> <p>In the current digital transformation, it is essential to immediately ensure the safety and stability of production to address the increasing complexity brought by the development of industrial automation and the diversification of system data transmission networks. A comprehensive tobacco enterprise safety risk early warning system based on big data analysis and advanced machine learning technology to provide scientific evidence for decision-makers. Develop an application capable of real-time danger detection and alerting; traditional methods have not yet addressed this flaw. The proposed system includes a scalable distributed data processing pipeline for collecting and normalizing data from environmental sensors, high-frequency sensors, and operator logs. In high-dimensional datasets affected by noise, novel feature engineering and dimensionality reduction methods are employed to extract key data. Boosted decision trees, kernel networks, and deep residual networks are important components of the ensemble learning system. The system's adaptability to anomalous events and the interpretable anomaly models can accurately identify changes in risk conditions. Identify instances of false negatives or missed reports in traditional monitoring systems; based on retrospective empirical evidence from actual production data, the accuracy of event occurrence predictions has significantly improved. The data-based modular early warning system has improved decision-making capabilities and operational safety in the tobacco industry environment. In order to expand its scope, we plan to promote it in more areas and enhance its preventive maintenance capabilities.</p>