Study On Forest Fire Prediction Based On Improved BP Neural Network

Forest fire is one of the greatest impacting and devastating natural disasters.As global warming,the probability of forest fire occurrence has increased,and caused huge damage on forest resources,natural environment and ecological balance.Not only has affected the development of social economy,but also endanger human's life and their property safety.Forest fire has the characteristics of sudden and randomness,and also have relationship with the meteorological factors at the same time.Forest fire prediction is an important premise of the construction on the forest fire prevention,the forecast of forest fire risk rating has great significance to guide the forest fire prevention.Meteorological factors are the important factors to affect forest fire,the forest fire risk rating stands for the probability of fire occurrence and its spread speed.According to the statistical yearbook of China,the frequency of forest fire,the affected forest area and the economic losses in JiangXi are located in the national top10 from 2004 to 2013,and the forest fires occur frequently.In this paper,I collected meteorological data of four stations in JiangXi(NanChang,Jing DeZhen,JiAn,GanZhou)from china meteorological data sharing network in October of 2013 to June of 2015,it contains 17 meteorological factors.And collected the forest fire risk rating data in this period of time from forest fire prevention network of JiangXi.To ensure the completeness and correctness of the experimental data,I selected 1017 groups of data as a sample in this paper.And then the correlation analysis was carried out on the sample data set by using the data processing software(SPSS),according to the analysis of the correlation between 17 meteorological factors and forest fire risk rating,I has chosen the 13 meteorological factors which have significant correlation with fire risk rating,then construct forest fire prediction model using BP neural network.In this paper,the BP neural network model was designed,and the optimal topology structure of BP was determined by the experimental trial,according to the result of the experiment,when the number of hidden layer neurons is 16,the prediction accuracy of model is highest which is 70.18%,so the topological structure of BP neural network is 13-16-1.In order to further increase the prediction accuracy of forest fire prediction model,we need to optimize the BP neural network.This paper use the genetic algorithm,particle swarm optimization(PSO)algorithm and PSO-GA algorithm to optimize the weights and threshold of BP,and build the corresponding fire prediction model,and using the experimental data to simulate.Then compare theprediction accuracy and the performance of the model.Experimental results show that,among the 4 prediction models,the BP neural network prediction model optimized by PSO-GA in this paper can improve the prediction accuracy,and the training error performance is well.