Deep Learning-based Fire Alarmer for Underground Power Cable Tunnel with Multiple Information Sources

Abstract

Regarding the issue of risk management and control of tunnel cable fires, a cable fire management and control scheme based on Gene Expression Programming (GEP) has been proposed. This scheme comprises three stages: firstly, a cable skin temperature rise model based on load variations has been established; secondly, a cable fire monitoring model based on YOLOv5 has been trained; and finally, a cable fire management and control scheme that integrates GEP has been proposed. Through the fusion analysis of data from the first two stages, the dynamic adjustment of the weight output of different parts is achieved to achieve automatic recognition of optimal states. Through experiments and analysis, the temperature variation patterns of cables under various loads and environmental conditions have been demonstrated, as well as the recognition performance of the fire monitoring model based on YOLOv5, and a dynamic planning scheme for the GEP model has been provided.