Research On Meteorological And Forest Fire Risk Forecasting Methods Based On Deep Learning

Meteorological factors are important factors that affect forest fires.Many experts and scholars use the meteorological factors to carry out the forecasting and forecasting of forest fire risks,so as to deploy related forest fire prevention and control work in advance.In the context of the complex global climate situation and the explosion of various meteorological data in the era of big data,it is difficult to achieve a more accurate prediction of forest fire risk with just-observed meteorological data.Accurate forecasting of relevant meteorological factors in advance can provide more timely and efficient data support and scientific research basis for forest fire forecasting.However,traditional meteorological and forest fire risk prediction models are mostly mathematical methods and shallow neural networks.These models are prone to modeling difficulties,overfitting and prediction accuracy when facing large amounts of meteorological data and unbalanced forest fire data sets.Problems such as low levels are not in line with the trend of the weather and forestry big data.In response to the above problems,based on deep learning related theories,this paper proposes models and methods for weather forecasting and forest fire risk forecasting respectively.The main contents are as follows:(1)Based on the Gated Recurrent Unit(GRU)network model,a maximum information coefficient(Maximal Information Coefficient,MIC)algorithm layer is added to the original GRU structure,so that the improved prediction model can be used for a variety of Correlation analysis of meteorological factors to screen out the input characteristics of the prediction target;using the Adam optimizer and R-Dropout method to optimize the parameters of the traditional GRU network and improve the model to further improve the prediction performance of the model.Finally,a MAR-GRU meteorological prediction model is proposed based on all the above improvements.The experimental results show that the prediction performance of the model proposed in this paper is significantly better than the original GRU network and other comparative models,which solves the poor performance of traditional meteorological prediction methods when facing large amounts of time series data.problem.At the same time,this part of the experimental results can also provide data storage for realizing continuous prediction of meteorology and forest fire risk.(2)Use meteorological factors as input data,forest fire risk level as prediction output,and select Deep Belief Network(DBN)as the basic prediction model.Aiming at the problem of unbalanced forest fire data set categories,a weighted oversampling SMOTE(Synthetic Minority Oversampling Technique)improved algorithm for forest fire data set is proposed.This algorithm is used to oversample a few fire samples,so that the whole sample set is in a balanced state.Finally,the improved SMOTE algorithm is integrated with DBN to construct a WS-DBN forest fire risk prediction model.The experimental results show that WS-DBN can significantly improve the prediction accuracy of traditional DBN,and wins in comparison experiments with various traditional forest fire prediction models.This result proves the superiority of the model proposed in this study to predict forest fire risk,and can provide a certain reference for the subsequent application of deep learning in other areas of forestry.(3)Integrate the two prediction models proposed in this paper to build a simulation system for continuous prediction of meteorology and forest fire risk.The system uses the output of the weather forecast as the input of the forest fire risk forecast,and finally displays the forest fire risk grades in the same period for forecast and display,so as to realize the continuous forecast from the weather to the forest fire risk and the visual display of the corresponding results.The system can provide a reference for the practical application of the two prediction methods proposed in this paper.