Research On Multi-feature Fusion Algorithm In Urban Functional Zoning And Anomaly Detection

With the development and maturity of sensor network technology,communication technology and positioning technology,the collected spatiotemporal big data has advantages such as wide coverage,high spatiotemporal resolution,and low cost compared with traditional data.The increase in urban population during the process of urbanization leads to an increase in the probability of emergencies such as car accidents and assemblies in urban areas.Serious consequences such as traffic jams may occur if timely detection and response measures are not obtained.And the technology of real-time identification of urban functional districts and anomaly detection using big data of time and space can not only help the government better plan and manage urban land,but also help maintain urban law and order.This thesis studies the application scenarios of urban functional zoning and abnormal event detection,and attempts to apply multi-data source feature fusion technology and the temporal and spatial neighbor anomaly detection framework to urban zoning and abnormal event detection scenarios,so as to provide support for city managers in urban planning and city stability management.Firstly,an improved method is proposed for multi-data source feature fusion for functional partition recognition in the context of urban functional partitioning based on spatiotemporal big data.Glove and LDA(Latent Dirichlet Allocation)models are used to extract the spatial location information,semantic frequency information and probability distribution information in POI(Point of Interest)data.The base station data is aggregated by the total number of people by the hour to obtain a time series vector.Multiple data sources are fused into one feature through the similarity matrix,and unsupervised clustering is used for recognition of urban functional areas.The proposed algorithm performs well on real data sets in terms of accuracy and consistency with land planning maps.Secondly,a method that combines multiple data source feature fusions and spatiotemporal neighbor anomalies is proposed to detect abnormal events.Because traditional anomaly detection methods usually only use a single data source,the accuracy of anomaly detection based on a single attribute is relatively low.In order to solve the above problem,the anomaly detection process framework proposed in this thesis is divided into two parts.The first part uses multi-data source feature fusion to solve the problem of singleness of the data source.The second part performs anomaly verification on the time and space neighbors of the abnormal points detected in the previous step to ensure the accuracy of the anomaly detection.The experiment on real abnormal events shows that the proposed anomaly detection framework has a good effect.