A Study On Urban Crowd Flow Prediction Based On Multi-source Data

Recently,the prediction of crowd flow,commuting flow,and traffic flow based on spatiotemporal data has become an important application in the field of urban planning,but the research in this direction tend to ignore the influence of the built environment in urban areas on the crowd flow.This thesis uses multi-source data to study the relationship between the built environment of urban areas and the crowd flow in the region,and build a prediction model,which is of great significance to urban planning.There are two main objectives of this study.First,obtaining the grid characteristics of the urban area by data preprocessing,this article models and predicts the expected crowd flow in the region.Secondly,by using the satellite map image data of the region and combining with the characteristics of the built environment,this article models and predicts the change trend of the regional people flow.Firstly,the thesis builds a data set based on multi-source data.Specifically,the signaling data is statistically processed and further gridded to quantify the label of the regional flow.Data such as buildings,land use,road network and satellite images are used to represent the characteristic attributes of the built environment of the corresponding region.In this thesis,appropriate feature selection and sample balance methods are used to effectively reduce the influence of the sparse feature attribute data and imbalanced label data.Second,the thesis argues that the regional population is influenced by its own built environment and its adjacent areas.In this article,the target urban area is meshized,and the entire urban topology can be regarded as a graph composed of nodes in the grid region,and the adjacency relationship between nodes can be regarded as an edge.Regional nodes are interrelated and influence each other through neighborhood relationship.Finally,the Graph Attention Network(GAT)is used to Embedding the characteristics,and a classification and regression prediction model is constructed by considering the sparsity of the sample features and the imbalance of the tags.Thirdly,the thesis introduces the satellite image data of the region,extracts the highdimensional features from the image data of the corresponding region by using Resnet50,and further fuses them with the features of the built environment to finally output the trend prediction model of the regional crowd flow.