Objective  IGiven the lack of methods for dynamic safety assessment related to disasters in drilling site environments for emergency rescue after underground space collapse, this study proposed a multisource data-driven methodology for dynamic safety assessment and real-time early warning.

Methods By comprehensively considering the disaster-inducing environment and disaster-causing factors at drilling sites for emergency rescue, this study established the safety assessment indices of the environment and their grading system. By integrating multi-source data on surrounding rock deformation, gas concentration, and drilling rig vibration, this study developed a safety assessment method through combination weighting achieved using the analytic hierarchy process (AHP) and entropy weight method (EWM). Moreover, this study constructed a multi-factor time series prediction model based on the autoregressive integrated moving average with an exogenous variable (ARIMAX) model and established a closed-loop mechanism consisting of monitoring, assessment, prediction, and early warning.

Results and Conclusions  The proposed safety assessment model enabled the dynamic optimization of weight allocation by integrating expert expertise and data distribution characteristics, significantly enhancing the stability of assessment results compared to individual subjective weighting models. This model yielded low prediction errors, with root mean square errors (RMSEs) of less than 0.12. A lightweight software platform was developed, significantly reducing warning response time by efficiently integrating the assessment and early warning methodology with visualization interfaces. Case validation demonstrated the high reliability of the proposed methodology. Overall, the proposed methodology overcomes the limitations of traditional static assessments, providing theoretical support and a technical tool for situational awareness and decision optimization in complex rescue environments.