| With the continuous development of social economy,the forms and means of crime are complex and varied,and the number of crimes is also increasing,which seriously affects social stability and people’s production life.Therefore,it is extremely important to grasp the law of spatial and temporal distribution of crime and its causal mechanism,and to achieve efficient and accurate crime prediction,in order to improve the efficiency of urban management and guarantee public safety.Currently,many scholars have conducted spatio-temporal analysis and prediction studies on crime behaviour.However,most of the existing prediction studies use a single machine learning or deep learning model that treats the temporal and spatial domains in an independent manner,often ignoring the spatio-temporal dependence of crime and thus making it difficult to obtain accurate prediction results.This paper takes the theft cases in the main urban area of Lanzhou City from 2016-2020 as the research object,and integrates various spatio-temporal analysis methods to explore the evolution of the spatio-temporal pattern of theft crimes;at the same time,it combines the common theories of crime geography to analyse the causal mechanism of theft crimes;on this basis,it proposes to use the GAERNN model combining graph self-coding and GRU to make spatio-temporal On the basis of this paper,we propose to use the GARERNN model combining graph self-coding and GRU to predict the future spatiotemporal distribution of burglary crimes in the main urban area,and verify the feasibility and effectiveness of the model in predicting urban burglary crimes.The research content and conclusions of this paper are as follows:(1)In terms of spatio-temporal pattern evolution: the spatio-temporal distribution characteristics of burglary crime are explored from the temporal,spatial and temporal perspectives respectively.At the temporal level,the SCII index was introduced,and "year","month","day" and "hour" were used as the time scales.At the spatial level,the global Moran’s I index and the local Moran’s I index were used to find a clear clustering pattern of theft crimes in the main city;At the spatio-temporal level,the spatio-temporal distribution pattern of theft crimes was explored using KDE and spatio-temporal hotspot analysis methods,and it was concluded that the number of hotspots of theft crimes in the main urban area of Lanzhou City from 2016 to 2020 was more than the number of coldspots,and the crime hotspot areas were mainly concentrated in the junction area of Qilihe District,Chengguan District and Anning District,while the crime coldspot areas were mainly concentrated in the eastern part of Xigu District and the southern part of Anning District.(2)In terms of influencing factors: Based on the theory of crime geography,the GWR model was constructed and the degree of influence of different influencing factors on theft crime was analysed,and the influence of each factor was derived from the following:entertainment facilities,population density,hotel and accommodation facilities,property and building facilities,and the distance to the nearest police station.(3)In terms of spatio-temporal crime prediction: visualisation of the prediction results of each model revealed that the visualisation results of the GAIRN model were most consistent with the actual data distribution;error analysis of each model revealed that the RMSE of the GAIRN model for each month was reduced by 1.02,3.58,1.29 and 0.45 respectively,compared to the MLP which had poorer prediction performance;The validity assessment of the submodule shows that the GAIRNN model reduces MAPE by 2.15%,10.07%,1.92% and 2.54%in each month compared to its variant GAE-LSTM.The experiments show that the GALENN model can significantly improve the spatio-temporal prediction accuracy of burglary crimes and can be used for active prevention and effective management of urban burglary crimes. |
