Construction And Application Of GIS-Based Model For Predicting Earthquake Casualties In Yunnan Province

Yunnan is located in the eastern margin of the collision zone between the Indian plate and the Eurasian plate.The whole area is cut by active faults,and the crustal activity is intense.The earthquake frequency is high,the earthquake magnitude is large,and the distribution is wide.In the aftermath of a devastating earthquake,transportation and communications are often disrupted,making it difficult for earthquake emergency services to quickly obtain information about the disaster.Therefore,rapid and high-precision casualty prediction is of great significance for emergency command and decision making,reducing disaster losses and strengthening people’s life and property security.In recent years,the use of Geographic Information System(GIS)technology to elaborate the interpretation of geological disasters,disaster background analysis and risk assessment is the focus of current research.Due to the complex and diverse characteristics of disaster factors,disaster environment and disaster bearing bodies,as well as the lack of earthquake disaster sample data,high precision earthquake disaster death toll prediction still faces great challenges.At present,relevant studies lack quantitative evaluation of high-dimensional features and the importance of features,and the prediction model can not represent the spatial relationship and geographical similarity relationship between samples,so it is difficult to build a fast earthquake death toll assessment system with high applicability.Based on this,the temporal and spatial distribution characteristics of destructive earthquakes in Yunnan Province since 1900 are analyzed.A large number of features were collected and extracted,and the importance of the features was quantitatively assessed.The graph neural network model was established to predict the number of earthquake deaths,and the accuracy was compared with other four traditional machine learning models.Finally,it was integrated into WebGIS system,aiming at improving the ability of rapid assessment of Yunnan earthquake disaster.The research content and main conclusions of this paper are as follows:1.Analysis of temporal and spatial distribution characteristics of earthquake disasters in Yunnan Province.In this paper,kernel density analysis,global Moreland index and local spatial autocorrelation are used to analyze the temporal and spatial distribution characteristics of destructive earthquakes in Yunnan since 1900.The results show that:(1)Since 1900,there have been five periods of seismic activity lasting 1-3 years,and the frequency of destructive earthquakes during active periods is 2-4 times of the average;(2)The magnitude distribution has a positive spatial correlation,and shows a significant aggregation trend;(3)From the seismic magnitude analysis,there are a large number of high values clustered in southwest Yunnan,high values distributed around low values and hot spots,and the cold spots mainly exist in northern Yunnan.2.Construction and evaluation of earthquake disaster data set.Firstly,the factors affecting the number of earthquake deaths were divided into four categories:disaster causing factor,disaster environment,disaster bearing body and disaster situation,including 73 samples and 64 characteristics.Then,a hybrid feature importance evaluation method is proposed,which takes Pilson correlation,Spearman correlation coefficient,information gain and average purity reduction into account.The results show that:(1)Among the top 10 characteristics of mixed importance,8 are the characteristics of population distribution,1 is the characteristics of geological disasters(the number of landslides),and 1 is the characteristics of damage degree(the highest intensity);(2)Feature dimension reduction based on the importance of mixed features can effectively improve the prediction accuracy of machine learning model,and the average R~2 is increased by0.083.3.Construction of machine learning model and accuracy evaluation.In this paper,a graph data construction method is proposed to construct the graph convolutional neural network(GCN)deep learning model.Meanwhile,particle swarm optimization support vector machine,BP-neural network,random forest and Bayesling machine learning models are constructed and trained and predicted on the data sets before and after dimension reduction respectively.The prediction effect of different models is analyzed,and the influence of dimension reduction based on the importance of feature mixing on model accuracy is analyzed.Accuracy verification shows that the R~2 accuracy of GCN model is 0.935,which is better than other traditional machine learning models.4.Realization of visual system based on WebGIS.Develop WebGIS system based on B/S architecture and Arc GIS API for Javascript,Arc GIS Enterprise,Django,Vue and other technologies.The main modules are monitoring and warning,earthquake disaster risk,intelligent emergency response and earthquake death toll prediction.To realize multi-faceted Yunnan earthquake information query statistics and interactive visualization,Yunnan earthquake casualties rapid assessment.