Research On Typhoon Disaster Risk Assessment And Disaster Loss Prediction

Guangxi Beibu Gulf Economic Zone is located in the northern part of the South China Sea.And the risk assessment and loss prediction of typhoon disaster here are studied in this paper.In terms of typhoon disaster risk assessment,geographical detectors are used to conduct single factor and factor interaction analysis to explore the driving mechanism of typhoon disaster based on the comprehensive disaster risk theory.Combined with the coefficient of variation method to determine the weight of participating factors,the spatial weighted superposition analysis was utilized to establish the typhoon disaster risk assessment model in the study area.In terms of disaster loss prediction,BP neural network model was improved by Genetic Algorithm to establish a typhoon disaster loss prediction model in the research area,so as to achieve accurate prediction and evaluation of future typhoon disaster losses.The main research results are as follows:(1)The temporal and spatial distribution of typhoons in the study area is analyzed on the basis of historical data.By collecting historical typhoon track data in the Northwest Pacific,this paper systematically analyzes the temporal and spatial variation of typhoons which affect Guangxi Beibu Gulf Economic Zone in the past 52 years.The results indicate that 81 typhoon disasters have occurred in the study area during the period,all from the Northwest Pacific,with an average of 1.5 typhoon disasters annually.In the past 20 years,the frequency of typhoons has increased slightly,instead of obviously.The path of typhoon that affects Guangxi Beibu Gulf region can be divided into five aspects:(1)Coastal landfall from Hainan to Zhanjiang of Guangdong--coastal landfall in Beibu Gulf--secondary landfall in Guangxi Beibu Gulf;(2)Pearl River Estuary to Zhanjiang coastal landing--westward into Guangxi Beibu Gulf Economic Zone;(3)Pearl River Estuary to Fujian coastal landing--westward drift into Guangxi Beibu Gulf Economic Zone.(4)Landing along the coast from southern Fujian to northern Guangdong--moving northwest to northern Guangdong--turning south to Guangxi Beibu Gulf Economic Zone;(5)Landing in northern Vietnam--turning southeast and entering southern Guangxi.The first three routes have the greatest impact on the area.(2)Based on the selection of factors that influence typhoon disaster,the factor detector and interactive detector in the geographical detector are resorted to identify the important driving factors that impact the disaster distribution in the area.The results demonstrate that the disaster factor is the decisive factor affecting the spatial differentiation,and there is no single factor affecting the distribution and change of typhoon disaster.(3)On the basis of the constructed typhoon disaster risk assessment model,the spatial distribution pattern of typhoon disaster risk in the area has been revealed by using the combination weighting method to determine the weights of each indicator.Finally,the typhoon disaster risk assessment model was established according to the spatial weighted superposition method.The basic spatial distribution pattern of typhoon disaster risk in the research area was considered as high risk in the southeast and low risk in the northwest.There are significant regional differences in the risk variation of typhoon disasters with the risk index gradually decreasing from the coast to the inland.(4)The prediction model of typhoon disaster loss in the area was established in accordance with BP neural network which was optimized by Genetic Algorithm.Input data were screened from three comprehensive layers,and namely,disaster causing factor,disaster bearing entities and disaster prevention and reduction capabilities.To take the direct economic losses of typhoon disasters over the past years as output data.BP neural network improved by Genetic Algorithm was used for nonlinear fitting.The example verification set test explains that the BP neural network promoted by Genetic Algorithm can reduce the error better,and the root-mean-square error is smaller than that of the single BP neural network,effectively improving the overall prediction accuracy.