Font Size: a A A

Research On Spatial And Temporal Distribution Prediction Model Of Theft Based On ConvLSTM Network

Posted on:2021-05-30Degree:MasterType:Thesis
Country:ChinaCandidate:S W GaoFull Text:PDF
GTID:2416330647958414Subject:Cartography and Geographic Information System
Abstract/Summary:PDF Full Text Request
Theft is a type of criminal case with the purpose of illegal possession.It has a high incidence in China,which seriously influences people's feeling of safety and blessedness.It has always been the most pressing mission for the police to perform.In the current process of urbanization in our country,a series of changes such as population transfer and space expansion have made the situation of theft police situation complicated and the situation grim.At the same time,theft are preventable crimes.The spatio-temporal distribution of theft is not random or uniform,but presents certain characteristics.By identifying and extracting the spatio-temporal distribution characteristics of theft police situation,predicting the distribution situation of the theft police situation based on hourly and community-level micro-scale.It is conducive to the police to more effectively distribute staff,forcing criminals to stop crimes due to the deterrent effect of the police,and to effectively combat such crimes.At present,many researches study crime prediction methods and models based on the spatio-temporal distribution,but the sparse distribution of crime caused by the microscopic spatio-temporal scale makes it difficult to extract the characteristics of the spatio-temporal distribution of crime,which is the main choke point facing crime prediction at the micro scale.At the same time,deep learning algorithms have developed rapidly and achieved breakthroughs in image feature recognition and longterm sequence memory.Therefore,the combination of deep learning algorithms in crime prediction methods and model research has become a development trend.This article takes theft as the research object,and analyzes the spatio-temporal distribution characteristics of theft and related environmental impact factors to build a model based on the Convolutional Long Short-Term Memory Network(Conv LSTM)to predicts the spatio-temporal distribution of theft based on the integration of hourly time periods.At the same time establishing a theft-prediction evaluation method to comprehensively evaluate the prediction results.The results of the paper mainly have the following three points.(1)By analyzing the spatio-temporal distribution of the theft in the study area,it is found that the spatio-temporal distribution of the theft in the study area has the characteristics of temporal periodicity and spatial aggregation,which proves the predictability of theft.On this basis,the spatial-temporal interaction of theft is explored,the near-repeat phenomenon of the theft police situation is analyzed and the effective time-space range is calculated.Considering the spatial-temporal distribution characteristics of the theft police situation,the appropriate observation object is determined.By analyzing the correlation between various environmental impact factors and theft,the environmental impact factors that are closely related to the theft in the study area are screened.It provides the basis for constructing a prediction model of theft spatiotemporal distribution based on Conv LSTM.(2)Through the four parts of adaptive spatial resolution method,Conv LSTM network design,related environmental fusion and prediction result evaluation method,a spatial-temporal distribution prediction model based on Conv LSTM network was constructed to solve the distribution characteristics of sparse theft.It solves the problem that the distribution characteristics of sparse theft are difficult to identify,and memorizes the dependence of the long-term sequence of the theft which realizes the prediction of the spatio-temporal distribution of the theft based on the hourly and community-level micro-spatial and temporal scales.(3)Through the analysis of the accuracy of the prediction model and the accuracy and aggregation of the prediction results of the spatio-temporal distribution,the verification analysis not only from the model fitting to the prediction results but also from the local to overall process is realized,which is used to evaluate the prediction of spatio-temporal distribution about the theft.The experiments show that the prediction model based on the Conv LSTM network can effectively realize the prediction of theft at the micro-spatial-temporal scale.The prediction result of the model is affected by the number of burglary alarms.The prediction effect will be better in the period when the number is larger.At the same time,the spatial-temporal distribution prediction model based on Conv LSTM network is compared with other prediction models based on deep learning.The results show that the prediction model constructed in this paper is superior to the CNN prediction model of theft and the FC-LSTM prediction model of theft.
Keywords/Search Tags:Theft, Spatio-temporal analysis, Near-repetition of crime phenomena, Crime prediction, ConvLSTM
PDF Full Text Request
Related items