Research On AOI Traffic Attribute Mining Based On Rider Trajectory Data

AOI(area of interest),that is,the region of interest,is a planar entity containing multiple POIs.It can be a community,school,park,etc.Urban AOI is the area with the most frequent human activities,which is closely related to people’s lives.In recent years,due to the spread of COVID-19,many AOI have been closed regulation,and with the change of local policies,the AOI’s property has also changed.Accurate and timely judgment of AOI traffic attributes is of great significance for urban management and research.It can also reflect the intensity of epidemic control and provide a reference basis for many decisions.Takeout riders have a wide range of activities,including narrow streets and communities,which are also the main carriers of crowd activities.Compared with traditional methods,mining AOI traffic attributes through takeout rider trajectory data has the advantages of high efficiency,low cost and high accuracy.Through literary research,the feasibility of this study is determined,and the scheme and technical framework of this study are determined.Firstly,it is determined that AOI traffic attributes include three categories: closed,no riding and passable.According to the nature and characteristics of different categories,the task is divided into two steps.In the first step,the GIS spatial analysis method is used to analyze the spatial distribution law of the track check-in points of AOI and the corresponding waybill.The two characteristics of the proportion of internal check-in points and the cluster number of internal check-in points are selected,and the appropriate threshold is selected through comparative experiments to screen out the closed AOI.The second step is to classify the internal trajectories of AOI,including riding and walking.Through the analysis of the characteristics of the trajectories,highly distinguishing features are extracted,and different machine learning models are used for training.The experiments show that the LightGBM model can achieve the best classification effect and high efficiency.Therefore,the LightGBM model and the parameters obtained through grid search are selected to predict the trajectories.By observing the proportion distribution of internal walking trajectories of different types of AOI,the threshold is selected to classify AOI.In addition,this study also attempts to use the end-to-end method based on deep learning to directly predict AOI categories.By converting the multi-source data from WGS84 coordinate system to web Mercator projection coordinate system,and then into image format,and finally superimpose the data in multiple channels.In terms of network structure,the experimental effects of different layers of networks and different structures of networks are compared,and the network with the best performance is selected.In addition,several groups of comparative experiments were carried out to select some important parameters and the best feature combination.After the training model converges,it is tested on the test set,and good classification results are also obtained.The AOI traffic attribute mining method based on rider trajectory proposed in this study can accurately classify AOI traffic attributes.At the same time,it has high real-time performance and broad application prospects.