Urban Functional Regions Recognition And Research Based On Data Mining

Land is the most important natural resource of a country.Land use analysis is of great significance for understanding the situation of the land.With the rapid development of technologies such as data mining and big data in recent years,in addition to traditional land-related data,taxi trajectory data is also very helpful in understanding urban land.The behavior of urban residents can reflect the interaction between urban land,which contains regular patterns of urban land use.Data mining through taxi data can provide a good understanding of urban land use.Besides,different urban land undertakes different social functions and has different social and economic attributes.Discovering social functional urban land has profound implications for understanding the spatial structure of the city and enhancing the understanding of urban land use.It has become a research hotspot in urban land use and urban planning.Therefore,in this paper,I perform multiple dimensional analysis on urban land use based on DiDi's taxi trajectory data of Chengdu in 2016,and propose a new model to discovering social functional urban land based on multi-source data.The main research work of this paper is as follows:(1)Data acquisition and preprocessing.The data covers central urban area and its surrounding areas of Chengdu,and the data contains POI data,road network data,DiDi's taxi trajectory data,urban land data and so on.The data preprocessing includes coordinate and format conversion,spatial analysis and processing,data cleaning,road network division,data labeling and segmentation.(2)Data analysis on urban land use based on DiDi's taxi trajectory data.This paper performs multiple dimensional analysis and visualization of time and space on different urban land,including residential land,commercial center,green space,office and public services,and compares the differences in the flow of people between different kinds of urban lands,which is helpful to understand the characteristics of different urban functional areas.Besides,this paper analyzes the supply and demand of some types of urban land from the perspective of taxi trajectories.(3)Research on urban functional area recognition model.Firstly this paper makes statistical analysis and numerical calculations on urban static data Then I select XGBoost and LightGBM as the base classifier to establish the urban functional area recognition model.And through stratified sampling to do 5 fold cross-validation,calculate the accuracy and recall rate of each type,and average accuracy to evaluate the model.Besides,I make statistical analysis and numerical calculations also on DiDi's taxi trajectory data and the application of doc2vec,LDA and other natural language processing techniques and Balanced Sampling oversampling technology to the recognition of urban functional areas have achieved good results.The experimental results show that the flow of people on urban land can reflect the characteristics of urban lands and the differences between urban lands.And models based on multi-source data(static data and dynamic data)are significantly better than models based on a single data source.What's more,the boosting tree model such as XGBoost and LightGBM is much stronger than the traditional model such as random forest as a machine learning base model.The experimental results also show that reasonable migration of doc2vec,LDA and other natural language processing techniques and Balanced Sampling oversampling technology can improve the average accuracy of urban functional area recognition models,and can also improve the classification category imbalance.