Research On The Early Identification Of Flash Flood Disaster Risk Based On Artificial Intelligence

Fujian province is located in the southeast coastal zone of China,which is located in the subtropical monsoon climate zone.With complex topography and numerous mountains,the hilly area accounts for about 90% of the province’s area,and nearly 80% of the 85 counties are located in mountainous areas.The hilly area has a large population.Due to the influence of monsoon climate and typhoon,and the special topography,flash flood disaster in Fujian province is frequent.Flash flood disaster not only causes the destruction of roads,Bridges,houses and other buildings,but also causes losses and threats to the safety of human life and property,which seriously restricts the rapid development of social economy in Fujian province.This paper analyzes the data of investigation and evaluation results of flash flood disaster in Fujian province,extracts the valuable information of flash flood disaster by using the method of natural language processing(NLP),constructs a structured flash flood disaster database,and lays a foundation for the research on the occurrence mechanism of flash flood disaster.Flash flood disaster evaluation index selection,and based on variable fuzzy recognition,BP neural network and random forest law construction of risk early recognition model,to evaluate risk of mountain flood disasters in Fujian province area,for the flash flood forecast warning department in flood control and disaster reduction and risk hierarchy to provide technical support,in this paper,the main research content is as follows:First,collect and sort out the flash flood disaster data of Fujian province and construct flash flood disaster database.Against mountain torrent disaster in Fujian province survey data is incomplete,the disaster events distribution chaos and valuable information is not to mention,natural language processing,hidden markov method is used to extract the mining disaster data in the Ming and Qing dynasties and flash flood disaster investigation achievements of effective data,using spatial overlay technology to improve disaster sites information,build a database of mountain flood disasters in Fujian province.It realizes data standardization and simplification,improves data quality and efficiency,facilitates search and analysis,and provides better service for mountain flood disaster analysis and evaluation.Secondly,based on the flash flood disaster database,the spatial and temporal distribution characteristics and genetic law of flash flood disaster in Fujian province were analyzed.For mountain flood disasters in Fujian province frequent problems,the use of mathematical statistics and visualization technology,from the space,time,topography,rainfall and other multi-angle analysis of the 3300 disaster sites in 1472-2019,research on flash flood disasters and the correlation between different factors,to explore the regularity of mountain flood disasters in Fujian province conditions,mining caused important influence factor of mountain flood disasters.The results show that the center of gravity of mountain flood disaster shifted from northwest to southeast in 600 years.The density of flash flood disaster has a positive exponential relation with rainfall and a negative exponential relation with slope.Third,the flash flood disaster risk index system and he mountain flood disaster risk assessment model are established.According to the distribution characteristics and genetic law of mountain flood disaster in Fujian province,the key index factors were selected to establish the index evaluation system.Three methods of variable fuzzy identification,BP neural network and random forest were used to construct the flash flood risk identification model in Fujian province.Based on the information of historical mountain flood disaster,the advantages and disadvantages of traditional model and machine learning model in modeling were analyzed.The results show that: in the index system,the correlation between rainfall,slope and elevation scale difference factors and early risk identification is higher than that of other disaster factors.In risk identification,the accuracy of variable fuzzy identification is low,but the model contains the value of subjective weight assignment,and the recognition of some high-risk areas is better than that of machine learning.In the case of complete training samples,the accuracy of random forest and BP neural network model is much higher than that of variable fuzzy recognition.Compared with BP neural network,the model accuracy and modeling flow have been greatly improved.Fourth,combining with the early risk identification model,the author drew the map of high-risk flash flood disaster in Fujian province,and analyzed the key prevention areas of mountain flood disaster.According to the disaster risk areas identified by the three models,the rationality of the areas was judged based on the historical flash flood disaster points and the actual hydrological information,and the regional distribution map of high risk was drawn based on various conditions,providing technical support for the flash flood disaster prevention department.The results showed that the high risk areas in the southeast and central coastal areas were distributed in a belt,the most southern coastal areas and the northwest were distributed in a circle,the southwest,the central and western regions and the north were distributed in irregular triangular points,and the disaster areas had obvious distribution regularity.