Research On The Status Change Of Urban Land Use Based On Taxi And POI Data

Land use classification is regarded as an important part of traffic planning.Planning decision makers can understand people's spatial distribution through rational land use classification.Changes in land use will also inform decision makers of changes in population distribution,thus affecting spatial planning.At the same time,land use classification is also an important tool for future urban planning expansion and traffic road planning prediction.Previously,due to the limitations of tools and methods,remote sensing images and formed urban planning and design maps were used as the basis of land use analysis in most cases.With the advent of the era of big data,data explosive growth,especially traffic data,it brings a sexual prospect to land use analysis.However,the diversity and heterogeneity of data,the extraction of useful data from large amounts of data,and the application of data to land use analysis have all brought challenges to us.Therefore,this paper studies how to analyze and forecast the land use situation through massive traffic data.The main work includes:(1)For massive data,single point data is partitioned by clustering algorithm.However,a single hierarchical clustering algorithm DBSCAN can't meet the needs of data clustering,because data points can't automatically determine the threshold,resulting in large or small data blocks.I-DBSCAN algorithm is introduced to improve the clustering,and the size of the polygon formed by the edge points of each clustering block is used to restrict the clustering blocks,so that the real clustering data can be obtained.Next,the code is implemented on the Hadoop distributed operating platform,which can greatly improve the speed of data processing.This is necessary to save time and computer resources for future research.(2)A new feature design method is proposed.After designing 12-dimension feature tables of data at two-hour intervals,each data is designed based on matrix eigenvalue to meet the requirements of machine learning.Then the land use situation is analyzed and predicted by machine learning method.Finally,the cross validation of ten folds is carried out.Experiments show that the accuracy of land use type prediction designed by this method reaches 89%.Then,through the analysis of the change of crowd flow,we find that the characteristics of people's travel in different periods of the day objectively reflect the real situation of land use change.(3)Classification of land use types by POI data,and data fusion by LDA text lexicon model combined with taxi data to correct misclassification and improve the reliability of the results.After data fusion analysis,the correct rate of correcting errors reached 89.3%.This research method plays an important role in promoting urban land use classification.