Abstract: Based on indoor experimental data, study the effects of wind speed, moisture content of fuel and size of flying fire source on the ignition probability, and explore the applicability of GA-BP neural network model to predict the ignition probability, to provide a theoretical basis for the study of forest flying fire behavior. Using the surface fuel of Larix gmelinii as the combustion material and the branches of Larix gmelinii as the flying fire source, the combustion experiment was conducted by adjusting the wind speed, moisture content of fuel and the size of the flying fire source. SPSS is used to analyze the influence of various factors on the ignition probability. Genetic algorithm is used to optimize the BP neural network, and GA-BP neural network model is established to predict the ignition probability. The model takes wind speed, moisture content of fuel and the size of the ignition source as input variables, and the ignition probability as output variables. The optimized number of neurons in the hidden layer is selected to train and predict the model, and the prediction ability is compared with the BP neural network model. Results showed that, the ignition probability has a positive correlation with the wind speed and a negative correlation with the moisture content of the fuel. The ignition probability of the large-diameter flying ignition source is significant; The single factor ignition ability from strong to weak is as follows: wind speed>size of fly ignition source>moisture content of fuel; The strong or weak ignition ability of multi-factor interaction is: wind speed and fly ignition source size>moisture content of combustible and flying fire source size>wind speed and moisture content of combustible>wind speed, moisture content of combustible and size of flying fire source; the prediction ability of the model is good, which is conducive to the prediction of forest fires.

Key words:  Larix gmelinii, litters, wind velocity, water content, ignition probability, GA-BP neural network