Research On Prediction Of Forest Fire Risk And Spread Based On LSTM

As an important part of the ecosystem,forests maintain the natural ecological balance and provide a large amount of resources for human production and their daily life.However,in recent years,with the rapid development of modern industry and the increasingly serious problem of global warming,the climate has become more arid and forest fires have become more frequent.Forest fires are a kind of natural disasters with high suddenness.They are destructive and difficult to deal with.Every year,forest fires bring us lots of human and economic losses.They have become one of the most important natural disasters threatening ecological balance,forest resources and human safety.Nowadays,it is of great practical significance to study the prediction methods of forest fire risk and forest fire spread.It can help us prevent and fight forest fires that have not occurred and those that have occurred initially,so we can avoid forest fires and protect the safety of people and property as well as forest resources.With the development of computer technology,deep learning has been widely applied in the field of prediction.The prediction of fire risk and fire spread has also made great progress on the basis of traditional empirical models.However,most of them are based on grid pixel classifiers.There is no in-depth study on the spatial clustering characteristics of forest fires,long-term trends and cyclical fluctuations in time series,and the coupling relationship between influencing factors.In view of the above problems,this paper aims to predict the fire risk and fire behavior based on the deep learning algorithm,and conducts in-depth research on the following aspects:1)Propose a forest fire risk prediction model based on buffer resampling:This article proposes a buffer resampling forest fire risk prediction model based on the design concept of LSTM.The model is designed based on the unique timestamp of time series,and buffer resampling method is used to enhance the training of forest fire aggregation characteristics for spatiotemporal nearest neighbor samples.In addition,Adam optimization algorithm is used to optimize the cross entropy loss function to further improve the prediction accuracy of the model.Finally,the performance of the model is verified through comparative experiments.2)Propose three progressive LSTM models to provide algorithmic basis for predicting forest fire spread:This article proposes three progressive LSTM models with different degrees of connection between main and secondary cell units,and conducts indoor ignition experiments to control other influencing factors such as terrain and combustibles.The model is trained using this dataset to learn the interaction between real-time wind speed and forest fire spread,and a prediction model for forest fire spread rate is constructed.And introduce the error circle center of gravity to analyze the accuracy of three models to confirm the above conjecture.3)Using CA-SDP model simulation training to predict real-time forest fire spread boundaries:We uses cellular automata combined with SDP algorithm to construct a CA-SDP model and train the law of forest fire spreading in eight directions.At the same time,modifications are made to other meteorological,terrain,vegetation,human activities and other influencing factors,and simulation experiments are conducted on wildfire data to predict real-time forest fire spread boundaries.Finally,analyze the consistency of forest fire spread rate,spread area,and final fire area to verify the performance of CA-SDP model in predicting forest fire spread.In summary,this article proposes three research methods for forest fire risk prediction and spread prediction: firstly,the forest fire risk prediction model based on buffer resampling(BRLS)can better capture the characteristics of forest fire aggregation;Secondly,the coupling relationship between forest fire spread rate and wind speed was verified through a progressive LSTM model,providing an algorithmic basis for predicting forest fire spread;Finally,using the CA-SDP model to simulate and train the prediction of real-time forest fire spread boundaries provides useful reference for forest fire prevention and control.