Characteristics Of Ignition Time And Fire Spread Of Solid Combustibles In Wildland Fire

Ecological civilization plays an important role in the development of the national economy.As one of the most frequently happened and difficult to put out disasters,wildland fire is the most dangeous natural disasters in the world.It’s a threaten for ecological civilization and forest resources,so it is necessary to strengthen wildland fire prevention force in the country that is short of forest.As one of the most important fire behaviors,fire spread determines the development of wildland fire.Study on the wildland fire science can help protect forest resources,build ecological civilization,improve emergency and wildland fire prevention management system.This paper mainly considers the time-varying radiative heat fluxes generated by wildland fire.The aim of this paper is to study and model pyrolysis and ignition characteristics of carbonized and non-carbonized combustibles under time-varying heat flux,and to further simulate and experimentally study the evolution of fire front.The main works are:1.Realizing the time-increasing heat flux in wildland fire.Pyrolysis and ignition time of combustibles under time-varying heat flux were systematically investigated.Analytical and numerical models of pyrolysis and ignition of wood and PMMA were established.Results show that the relationshipbetween ignition time and parameters of time-varying heat flux can be described by a power function.In the experiments,we used aproportional-integral-derivative(PID)method to generate linear and square increasing heat flux based on the laboratory’s early stage of fire platform.Pyrolysis and ignition time of wood and PMMA irradiated by the heat flux were measured,validatingtheoretical analysis.2.Since pyrolysis,heat transfer and other factors were neglected in the analytical solution,we further adopted machine learning to generate a data training prediction network through numerical models in order to quickly and accurately obtain the ignition time of combustibles of various parameters in different environments.The results indicates that machine-learning-based method is more accurate than analytical solution and faster than numerical solution,providing new ideas for further wide applications of machine learning methods in pyrolysis research.3.Short-term prediction of fire propagation was achieved based on data fusion.Parameters in the empirical model of rate of spread can be modified by real-time observation data so that the model can instantly respond to parameter changes caused by time and spatial changes.The model included three modules:real-time velocity field measurement,fire propagation front evolution prediction,and velocity field parameter modification.The velocity field measurement module obtainednormal velocity field at the fire front by extracting real-time front position and calculating its displacement in the normal direction.The fire propagation frontal evolution prediction based on level set method evolutes the fire front at a given time in the given velocity field.Since fire front is hidden in a high dimensional space,level set method can conveniently handle the topological change such as front intersection and disappearance.The velocity field parameter correction module corrects the parameters in the machine learning prediction network based on measurement.Three modules were then individually tested.Firstly,influences of wind speed,slope,front intersection and other factors on the fire spread were investigated based on the combinationof level set method and empirical formula of velocity field.It was shown that the level set method can effectively obtain the front evolutionin a given velocity field.Secondly,velocity field normal to fire front numrically generated by level-set method was calculated and compared with known value for validation.Thirdly,neural network predicting velocity field was updated utilizing the velocity measurement,with which fire front evolution was predicted and compared with the known fire front,demonstrating that our model can avoid explicit expressions of empirical formula of rate of spread and effectively predict fire front in different environments.4.The model was finally tested with experimental shrubland and leaf fire propagation.Machine learning method was employed to update the velocity field with real-time measurement of rate of spread(RoS).Short-term prediction of fire perimeter was then compared with observed fire perimeter,demonstrating that our method can achieve fire propagation without any empirical RoS formula.The work provides theoretical foundations for application of data fusion in a fire spread model.