Research On The Spatial-Temporal Crime Prediction And Dynamic Assistant Decision-Making Method

In the current social background of police resources limitation or insufficiency,it becomes one of the most important questions that how to provide dynamic decision-making support for the optimal allocation of police resources by accurately integrating the spatial-temporal crime prediction technology in the study area of predictive policing mode in terms of public security risk prevention and controlling.In response to the above question,an integrated model（STCP-RLA）based on Res Net-LSTM and attention mechanism for spatial-temporal crime prediction is proposed in this study.The STCP-RLA model can extract the spatial-temporal features of crime incidents numbers and combine them with the external features.Then,a multi-agent simulation method and system for crime risk prevention and controlling in urban areas is established based on the social force model.And this system is used to study the crime risk compensation effect of police resource allocation.Moreover,a dynamic decision-making method for crime risk prevention and controlling is proposed based on the Bayesian network.Sensitivity analysis is conduced to examine the key nodes of the network.And scenario analysis is used to validate the method of decision-making in terms of police resource allocation.The main conclusions of this study are as follows.（1）The effectiveness of the STCP-RLA model is verified by analyzing daily theft and assault data within 22 police districts in Chicago,US from 1 January 2015 to 7 January 2020.This research finds that the spatial-temporal distribution of crimes predicted by the model is close to actual results,and the evaluation indicators MAE and RMSE are both lower than 2incidents per day in the test set of theft and assault incidents,which are lower than the other 8comparison models such as Conv-LSTM.In addition,the STCP-RLA model provides a framework for synchronous computation of temporal and spatial-temporal predictions by fusing features through attention mechanism,thus the prediction with high timeliness is achieved under the premise of high precision.（2）The law of crime risk compensation is studied by using the multi-agent simulation system of crime risk prevention and controlling in urban areas.Taking the first police district of Chicago as the research object,this study analyzes two resources,namely the police and surveillance points,and then investigates their relationships,in terms of quantity and distribution,with crime risk respectively.The research results show that allocating police resources based on the distribution of crime hotspots can be more effective in risk prevention and controlling.Additionally,with the increase in police resources,the marginal utility decreases.Moreover,the single exponential function can better fit the relationship among the police force,monitoring points and the number of criminal incidents,and the R² is 0.974 and0.957,respectively.（3）Taken the first police district of Chicago as an example,the proposed Bayesian network dynamic assistant decision-making model for crime risk prevention and controlling is used to analyze scenario.The results indicate that when the number of predicted cases is"high",it is necessary to simultaneously implement"increase patrol spatial-temporal coverage scale to 2times larger than the current scale"and the other three alternative police resource strategies provided in this study with a probability over 40%to realize keeping the actual crime risk at a"low"level.When the expected input of police resource is limited,the model can produce an optimized mixed resource allocation plan,and the increase in the expected value of crime risk does not exceed 2.5%.Among the four policing strategies,the"benefit/cost"ratio of"the current scale of video surveillance time-space coverage is doubled"is the highest,which is3.564×10^-6.The Bayesian network dynamic assistant decision-making method driven by crime prediction as developed in this study is expected to provide method support for several scenarios such as risk prevention and control in public safety,police force resource allocation,community policing,and large-scale event security.