Application of Grid Management in Spatio-temporal Prediction of Crime

For traditional data-driven modeling method is not applicable to predict possibility of a crime under the complex inner link problems, put forward a kind of based on grid management concept, a new method of crime prediction system optimization modeling, the method focuses on the grid of geographical features, implementation of crime time, location, weather data, the network platform of data characteristics, such as the existence of the internal correlation analysis. This article in the Chicago area as the research object, by using the BP neural network algorithm modeling, and crime in the pretreatment process data into three-dimensional space-time grid model, establish units within the grid and relations between adjacent grid data, according to the grid position and distance to determine the influence weight, according to the weight integrate the input data set as a model. The experimental results show that compared with the traditional meshless modeling effect, the MAE error of the four types of cases on the
$50^{\ast}50$
grid is reduced by about 0.03 on average, and the RMSE error on the
$50^{\ast}50$
grid is reduced by about 0.09 on average. The research proves that the selection of appropriate algorithm on the basis of reasonable application of meshless management has a better effect.