A physics-informed artificial fish swarm algorithm for multiple tunnel fire source locations prediction

Abstract

To address multiple tunnel fire source locations prediction for better firefighting, this work aims to establish a model-driven and data-driven fusion algorithm, which is named as the physics-informed artificial fish swarm algorithm. In the algorithm, the artificial fish swarm algorithm is utilized to avoid complex fire mechanisms modelling, which makes the algorithm owns the advantage of the data-driven methods. Meanwhile, a physical model of the longitudinal ceiling temperature distribution relative to fire source locations and some physical mechanisms are utilized to guide the data-driven artificial fish swarm algorithm for converging to the reasonable prediction, which makes the algorithm owns the advantage of the model-driven methods. Two full-scale tunnel fire experiments with two pool fire sources and three pool fire sources as well as one full-scale cable fire experiment in the tunnel are utilized to support and verify the algorithm, which can be utilized to achieve multiple tunnel fire sources prediction and ceiling temperature distribution prediction in tunnel fires.