Hybrid Feature Fusion-Based High-Sensitivity Fire Detection and Early Warning for Intelligent Building Systems.

Area of Science:

• Building Safety Engineering
• Artificial Intelligence
• Sensor Technology

Background:

• Early fire detection is crucial for intelligent building safety but challenging due to subtle initial combustion signals.
• Existing methods often struggle with the small changes and fluctuations in environmental parameters during early fire stages.

Purpose of the Study:

• To develop a high-sensitivity early fire detection and warning system for in-building environments.
• To address the limitations of current methods in detecting fires during their initial combustion phase.

Main Methods:

• A hybrid feature fusion approach combining temperature, smoke, and carbon monoxide concentrations.
• Optimization of a Backpropagation Neural Network (BPNN) using a Genetic Algorithm (GA) and a Least Squares Support Vector Machine (LSSVM) using Particle Swarm Optimization (PSO).
• Fusion of optimized model outputs using D-S evidence theory for final decision-making.

Main Results:

• The proposed method achieved over 96% accuracy in detecting various early fire types, including polyurethane foam, alcohol, beech wood smolder, and cotton fabric smolder.
• Experimental validation demonstrated the system's effectiveness in real-world scenarios.
• The hybrid fusion approach significantly improved detection sensitivity and reliability.

Conclusions:

• The developed hybrid feature fusion method offers a highly sensitive and reliable solution for early fire detection in buildings.
• This approach enhances intelligent building safety by providing timely and accurate fire warnings.
• The integration of optimized AI models and sensor data fusion represents a significant advancement in fire safety technology.