A multi-neural network fusion algorithm for fire warning in tunnels

Abstract

Fire early warning plays a key role in fire-fighting actions for alleviating the potential casualties of the trapped personnel and firefighters. This paper targets to achieve a fast detecting whether fire occurs or not in urban utility tunnels during fire initial stage by using the data-driven artificial intelligence algorithm. To improve the correct forecast rate, an auto-enhanced multi-trend back propagation neural network algorithm is developed based on the two betterments. One is that many neural networks are constructed to a strong multi-neural network for achieving enhanced-fusion prediction. Another is that one special trend extraction part is coupled in the algorithm to enhance the perception of no obvious temperature variation trend during fire initial stage. Finally, two kinds of full-scale tunnel fire tests were conducted to support the ability and effectiveness of the algorithm. The algorithm can successfully detect whether fire occurs or not with a 96.8% correct forecast rate by learning the temperature data during fire initial stage, even if the maximal range of temperature change is less than 3°C. In addition, to achieve such a high correct forecast rate, only one sensor is necessary for fire early warning in 60 m long tunnel by using the algorithm.