Research On Forest Fire Spread Prediction Based On Long Short-term Memory Networks

Forest fire is a kind of natural disaster with great hazard,which is affected by many factors such as the nature of combustible materials,terrain conditions,climatic factors and so on.Ground fire spread is the primary stage of forest fires and the intermediate process of transition to higher intensity forest fires.It is of great practical significance to study its occurrence under different environmental conditions.Fire spread rate is the most important physical quantity that characterizes the spread of fire.It will interact with the local environment during the development of ground fire spread.This article focuses on using the mutual influence of fire spread rate and local wind speed to accurately predict the spread of ground fire.Using the long-and short-term memory neural network to capture the dynamic characteristics of time series,three progressive structures of LSTM neural networks are proposed to model the mutual influence of the fire spread rate and wind speed in the process of forest fire spreading.The terrain,combustibles and environmental factors are controlled to design indoor burning experiments,and the data set is used to train and analyze the model’s fitting ability and prediction accuracy.The error center of gravity distribution map is used to evaluate the accuracy of the three models.It is found that the FNU-LSTM network based on the hypothesis that there is a strong interaction between the wind speed and the fire spread rate during the fire spreading process can achieve better performance.On this basis,the wildfire data set is used to compare and train the FNU-LSTM model proposed in the article and the excellent LSTM improved models involved in other papers to verify the applicability of the model in different environments.In order to fully predict the position of the fire front and the fire area of the forest fire spread,the cellular automata equipped with a progressive long and short-term memory neural network-FNU-LSTM model is used to train and predict the fire spread rate and wind speed in eight directions,combined with the wind direction,Slope,aspect,and vegetation coverage are timely revised to the output of the model to make it more in line with the specific environmental factors of the fire site.Using the three wildfires’ data and the field burning experiment data collected by drones,the simulation results of the model are used to analyze the accuracy of the fire line position and the consistency analysis of the fire area.The results show that the CA-FNU model designed in this paper has the high accuracy to predict the spread of forest fires.