A Self-Developed Fire Early Warning System Based on Gas Detection and Graph Convolution Calculation Method

Abstract

This study develops an artificial olfactory system for the early monitoring of fire risk in electric cabinets. Compared with existing fire detection methods such as temperature, smoke, sound, and current, the detection object of artificial olfactory sensors is abnormal odor, the diffusion of odor does not consider the complex structure of the electrical cabinet, and the detection results do not need to distinguish variable electrical conditions. In the study, we develop an artificial olfactory training device equipped with a sensory data collector to collect odor information from six combustible materials under smoke-free conditions. Based on the designed fast Pearson graph convolutional network (FPGCN), volatile gases from different overheated materials are identified with high performance under different heating times (at 1–350 s, an accuracy of 98.08%, a precision of 98.21%, and a recall of 98.01% are achieved), which proves the feasibility of the artificial olfactory training device.