Research On BPNN-DIOC Network Optimization Theory And Its Forest Fire Risk Prediction

Forests,one of China's important resources,can not only protect the environment and balance ecosystems,but also reduce the impact of climate change on people.However,the occurrence of forest fires seriously affects the development of forest resources,which not only causes the passage of ecological resources,but also poses a threat to people's economic property and even life security.Therefore,accurately predicting the occurrence of forest fires is an important means to reduce the forest fire hazards.In this paper,artificial neural network is used for forest fire risk prediction research.For the traditional BP neural network,the prediction accuracy is low and is easy to fall into local extremum.This paper modifies the network from the BP neural network structure and initializing network parameters respectively in order to improve the precision of network prediction.Specific tasks include:1.Improving the BP network structure.As we all know,BP neural network can achieve non-linear mapping between input and output,however,if the actual input and output have a certain linear relationship besides non-linear relationship,using BP neural network may not get an accurate relationship.In view of this situation,this paper proposes an improved network structure-Direct BP neuralnetwork(BPNN--DIOC).This network adds a linear connection unit between the input layer and the output layer of the BP neural network.To verify the impact of input-output connections on network generalization capabilities,the paper uses the electricity load data from AEMO NSW in September 2015 as an example.Example prediction results show that the direct BP neural network can not only improve the prediction accuracy of the network compared with the BP network,but also can reduce the number of neurons needed in the hidden layer to reduce the number of parameters that the network needs to adjust during training.2.Optimize network initial connection parameters.In the network initialization stage,the BP algorithm randomly generates network weights and thresholds,which can easily cause the decrease of network convergence accuracy and falling into local extremes.In this paper,fruit fly optimization algorithm FOA,which has just been proposed in recent years,is used instead of the BP algorithm to initialize the connection weights and thresholds of the network.For the deficiencies in the basic FOA algorithm,this paper proposes an improved fruit fly optimization algorithm(IFOA)with dynamic search steps,and four standard test functions are used to analyze the optimization capabilities of IFOA,and the results show that IFOA has a strong overall optimization ability.Finally,the IFOA algorithm is combined with BPNN-DIOC,which is based on IFOA to optimize the initial connection parameters of the BPNN-DIOC network.The simulation results show that the IFOA-BPNN-DIOCconstructed in this paper provides a more effective framework for the prediction model.According to the previous improvement on the BP neural network,this paper uses the IFOA--BPNN-DIOC network for forest fire risk prediction reseach of forest fire risk forecasting in two aspects.The first is to predict the occurrence of forest fire based on meteorological factors,the second isspatial interpolation prediction of forest fire meteorological factors to improve the spatial resolution of meteorological data.Specific tasks include:1.In the first applied research,four meteorological factors:temperature,relative humidity,wind speed,and precipitation are taken as the input of the network,the output is whether there is forest fire.Taking the records of Taiyuan City in 2011 and 2013 and Guilin City's 2005 and 2010 meteorological data and forest fire occurrence records as research objects,a forest fire occurrence prediction model based on IFOA-BPNN-DIOC was constructed respectively and compared with other models.The simulation results show that the IFOA-BPNN-DIOC network has the highest prediction accuracy,and this model is not only effective in regions where forest fires occur more frequently,and it also has good prediction effect on the area with less forest fire.Which illustrates it has good versatility.2.In the second applied research,the longitude,latitude,and altitude of the geographic location of the meteorological station are taken as the input of the network,the daily average temperature actually observed is used as theoutput.Taking the data of the meteorological station in Shanxi Province as an example,and a forest fire temperature spatial interpolation model based on IFOA-BPNN-DIOC is constructed.The results of IFOA-BPNN-DIOC spatial interpolation model are compared with those of IFOA-BPNN,FOA-BPNN-DIOC and FOA-BPNN networks.The RMSE of the four network interpolation results are 0.0716,0.0874,0.0815,and 0.0908 respectively,which fully illustrates that BPNN-DIOC has higher prediction accuracy than BPNN network,and IFOA has better over optimization capability than FOA algorithm.