Research On Forest Fire Warning And Emergency Management Models Based On Machine Learning

To mitigate the damage caused by forest fires,it is essential to study the influence of various forest environmental factors on these fires,and to predict both the risk of fire occurrence and the potential locations where fires may cease.The spatial distribution pattern of forest fires should be summarized,and the risk levels of forest fires should be categorized across different regions.By predicting possible fire boundary locations,we can provide decision-making support for forest fire prevention and control departments.After a fire occurs,it is crucial to plan rational evacuation routes to minimize loss of life and property.Existing forest fire models play a certain role in predicting and preventing forest fires.However,with the increased capability to gather forest fire data,the accuracy of forest fire risk prediction will be further improved.Therefore,in order to find a superior forest fire model,this paper focuses on the northern forest region of Yanyuan County in the Liangshan Yi Autonomous Prefecture of Sichuan Province.Based on machine learning techniques and path-search algorithms,we conduct research and analysis on the forest fires in the area under study.The specific research contents and results are as follows:(1)Utilizing products such as Landsat-8,the Moderate Resolution Imaging Spectroradiometer(MODIS),and Digital Elevation Model(DEM),we extracted data on fire points in the forest during months of high fire occurrence,as well as data on factors in the forest environment such as topography,climate,vegetation,and human activities.From this,we have established a Geographic Information System(GIS)database for forest fires in the study area.(2)Forest Fire Risk Prediction.We selected eleven factors impacting forest fires as driving factors,including slope,aspect,altitude,topographic wetness index,distance to waterways,temperature,precipitation,water vapor pressure,normalized difference vegetation index,distance to populated areas,and distance to roads.We conducted statistical analyses on the correlation between each driving factor and between the driving factors and the occurrence of forest fires.We constructed a forest fire prediction model based on the Light Gradient Boosting Machine(Light GBM)method and used the Random Forest(RF)method for comparison.According to the results of the F1-score,accuracy,and Area Under the Curve(AUC)model evaluation metrics,the performance of the Light GBM model was superior.(3)Forest Fire Boundary Prediction.Given the complexity and diversity of the spatial aspects of forest fire environments,we divided the fire environment into six types of fire environment models according to different sampling distances.We established a Matched Case-Control Conditional Light Gradient Boosting Machine(MCC CLight GBM)hybrid machine learning model to predict the positions where fires would stop under the six types of fire environments and to generate predictive maps of fire boundary formation probabilities.We then carried out performance analysis of the prediction model under different environmental models using F1-score,accuracy,and AUC model evaluation metrics.We also set up a Matched Case-Control Conditional Random Forest(MCC CRF)for comparison.The results showed that the MCC CLight GBM prediction model for the fire environment model with a case 0 m and control-120 m to-480 m had the highest efficiency.The experimental results indicated that fires in this area are likely to stop near roads,settlements,and areas with significant topographical variations.(4)Escape Route Planning.Taking into account the impact of terrain and vegetation on the speed of evacuees,and combined with the risk of fire occurrence,we proposed an improved A*algorithm to establish an evacuation path model.This model plans a rational evacuation route that allows evacuees to reach the target area in a relatively shorter time.We then overlaid this onto a topographic relief map of the study area.The results showed that most of the planned routes coincide with areas of minimal topographic variation.Compared to the paths searched by the traditional A* algorithm,although these routes are longer,they take less time for evacuation and are further away from high-risk fire areas.