Urban Functional Area Identification Based On POI And GPS Trajectory

In the long-term development of the city,different functional areas have been formed by human design or natural reasons,and bear different urban functions,such as commercial areas,residential areas and entertainment and leisure areas.Different functional areas often have different activity patterns.Understanding the distribution of urban functional areas is conducive to making reasonable evaluation and adjustment of urban layout and alleviating traffic pressure.With the rapid development and application of Global Positioning System and International Comminication Technology,various geographic big data appear in the public view,such as Points-of-Interest,mobile signaling data,floating car data,check-in data and so on.The multi-source,massive,current,high spatial and temporal resolution geographic big data contains rich information such as human behavior patterns,urban structure,and traffic operation status.Compared with the traditional low efficiency,long working cycle,heavy workload of field research or remote sensing image processing.The analysis method based on geographic big data can comprehensively,efficiently and accurately reflect the development and changes of urban spatial structure.The mining of geographic big data has become an important means of urban perception and intelligent planning.The purpose of this paper is to mine the context relationship of different kinds of interest points in geospatial space based on GloVe word embedding model,and use the nearest distance principle to construct functional area corpus to obtain word vectors.Based on the K-Means method,the urban traffic areas are classified on the classification scale of Traffic Analysis Zones.When obtaining the word vector of the traffic area,the area and activity intensity attributes of POI are used as weighting factors to weight and sum different types of POI to improve the classification accuracy of urban functional areas.After experimental analysis,this paper draws the following conclusions:(1)This paper uses the same corpus construction method and parameters,based on the word2 vec model to obtain the word vector of the POI category,and compares the model classification results with the GloVe model classification results.The results show that the accuracy of the experimental method is 90 %,which is significantly higher than the word2 vec model().(2)When constructing the corpus of GloVe model,the number of nearest POIs has a significant impact on the quality of word vectors.When the number of neighbors is limited to a small range,high-quality word vectors can often be obtained.The word vector dimension has an important influence on the experimental results.The higher the dimension,the higher the classification accuracy of different types of functional areas.(3)The introduction of area weighting factor can significantly improve the recognition accuracy of scenic spots,medical services and industrial land.Reduce the entropy of mixed functional areas.(4)This paper combines the GloVe model with the widely distributed POI data in the city,and makes full use of the spatial context relationship between different types of POI to classify the urban functional areas of Chengdu.It adds new theoretical knowledge and methods to the identification function of urban spatial structure.The research results can provide reference for the study of urban spatial structure.This paper proves the superiority of GloVe model in obtaining word vectors,and analyzes the optimization of model parameters.Fully consider the impact of area and activity intensity on the classification results.It provides a reference for the identification of urban functional areas.