A study on risk assessment and system development of tunnel lighting facilities based on XGBoost and Bayesian...

Tunnel lighting facilities exhibit significantly increased failure rates and accident risks during long-term operation and maintenance (O&M) due to environmental erosion and equipment aging. Addressing the high latency and resource inefficiency inherent in traditional periodic maintenance models, this study proposes a data-driven dynamic risk assessment methodology for risk assessment and system development of tunnel lighting facilities, with a specific focus on leveraging XGBoost and Bayesian optimization. The research process encompassed several steps: (1) multi-source data fusion for risk factors: integrating 12 critical indicators, including illuminance levels, waterproofing status of electromagnetic contactors, internal tunnel temperature, voltage stability, relative humidity, pole integrity, tunnel wind speed, cumulative operational days of lamps, number of damaged lamps, line corrosion, luminaire damage status, and cumulative operating hours of light sources-to construct multi-dimensional feature engineering. (2) Development of a risk assessment model: utilizing the XGBoost algorithm enhanced by Bayesian Optimization (BO) for automated hyperparameter tuning, effectively capturing nonlinear and temporal features while overcoming the inefficiency of traditional hyperparameter tuning methods. (3) Model validation and lightweight system development: Validated using empirical data from Caoguo Mountain Tunnel, this study innovatively establishes a closed-loop workflow integrating "model training, dynamic assessment, and risk prediction". An embedded assessment system, developed in PyCharm, implements an optimized three-tier architecture ("data acquisition-algorithmic computation-decision output"). This system achieved a 98.0% failure prediction accuracy on the Caoguo Mountain Tunnel dataset, thereby demonstrating a significant reduction in the operational and maintenance (O&M) risks of lighting facilities. These findings provide scientific theoretical and practical foundations for tunnel infrastructure maintenance.