A Tunnel Fire Detection Method Based on an Improved Dempster-Shafer Evidence Theory

Area of Science:

• Engineering
• Computer Science
• Safety Science

Background:

• Tunnel fires pose significant safety risks, necessitating reliable detection systems.
• Existing multi-sensor fire detection methods often suffer from data ambiguity and inconsistency.
• The Dempster-Shafer (DS) evidence theory offers a framework for handling uncertainty but can face evidence conflict issues.

Purpose of the Study:

• To propose an improved Dempster-Shafer (DS) evidence theory for robust multi-sensor data fusion in tunnel fire detection.
• To address and resolve evidence conflict problems inherent in the classical DS theory.
• To enhance the accuracy and speed of tunnel fire detection for improved personnel safety.

Main Methods:

• A two-level multi-sensor data fusion framework was developed.
• The first level involved feature fusion of same-type sensor data and basic probability assignment (BPA) calculation.
• The second level optimized BPA by calculating and normalizing evidence conflict, followed by integration using classical DS theory.

Main Results:

• The improved DS evidence theory achieved detection probabilities ranging from 67.5% to 84.1% across six simulated fire scenarios.
• Fire occurrence was identified in approximately 2.4 seconds.
• The method demonstrated a significant improvement in detection accuracy (64.7% to 70%) compared to traditional approaches.

Conclusions:

• The proposed improved DS evidence theory provides a feasible and superior method for multi-sensor data fusion in tunnel fire detection.
• The enhanced accuracy and speed of detection are crucial for ensuring personnel safety in tunnel environments.
• This approach effectively mitigates issues of data ambiguity and evidence conflict in sensor networks.