A model-driven and data-driven fusion prediction algorithm for the fire source location of an underground pipe gallery based on ant colony optimization

Abstract Most of the fire theoretical models were established by simplifying the complex burning process based on the specific fire experimental data. They are typically applicable to similar specific fire scenes, which severely restricts their application scope for complex real fire disasters. To overcome the deficiency of theoretical model-driven methods, a data-driven smart inverse algorithm is developed for predicting the fire source location and temperature field of the underground pipe gallery. The developed algorithm merely relies on the monitored temperature data at certain sensor locations and a simple model, which pertains to a backstepping technique from the results to the causes. As a case study, the fire source location and temperature field of an underground pipe gallery were predicted and compared with the corresponding experimental results. The distance between the predicted fire source location and the actual fire source is less than 0.5 m. The maximum error between the predicted temperature and the monitored value is less than 13%. The mean absolute error is about 3.6%. Model parameter analysis and the effect of the discretization of the searching space were further discussed. The results validate the capability of the algorithm.