A Matrix Factorization Fusion Model For Urban Functional Region Identification

Urban functional regions(UFR)identification is significant to urban planning and sustainable development.The accuracy depends to a large extent on the information implied behind the data.Currently many methods use land data,POI data and social media data for identification.However,the formation of UFR is a process in which various activities are related,and the accuracy of identification using only single data is low.Data fusion becomes an effective method to improve the identification accuracy.However,traditional fusion methods ignore the shared semantics among mule-data sources and do not consider the similarity between data,so it is still challenging to accurately identify urban functional regions.To address the above problems,this paper selects POI data and trajectory OD data within the fifth ring area of Beijing,and extracts trajectory OD interaction data and trajectory OD timefrequency data from trajectory OD data,and proposes a data fusion method based on context coupling matrix factorization considering context relationships to identify urban functional regions by using the potential complementary relationships existing between the three data.The research results can provide scientific decision support for urban planning.The main research contents of this paper are as follows:(1)A fusion model based on contextual coupling matrix factorization is proposed.With the aim of improving the accuracy of urban functional area identification,this paper investigates and reviews the existing methods and practices for urban functional area identification,and proposes a data fusion model based on contextual coupling matrix decomposition.The identification accuracy is improved by mining the potential complementary information between data.First,the model extracts potential information from source data decomposition and stores it in the form of a matrix.Secondly,multiple matrices are decomposed with cooperative coupling by fully considering the contextual relationship between data.Finally,oriented to the problem of good or bad fusion effect,the optimal solution is calculated by using gradient descent method to realize data fusion,and the fused data is used for functional area recognition to improve the accuracy.This study applies the idea of matrix decomposition to identify urban functional areas,which provides a new idea for urban functional area identification.(2)A functional region identification framework based on context coupling matrix factorization fusing multi-source data is constructed.The study of urban functional region identification plays an important role in optimizing urban spatial layout.However,existing technical approaches and application practices often use multiple sources of data to improve recognition accuracy while ignoring the advantages of data coupling and information complementarity.In view of this problem,this paper investigates how to improve recognition accuracy from the perspective of potential correlation between data.Therefore,we propose a fusion framework for identifying urban functional areas,aiming to explore the analysis model applicable to urban functional areas,so as to investigate a recognition framework with high recognition accuracy and high practicality.(3)Case analysis of identification frameworkAn example study is conducted in the 5th Ring Road area of Beijing,and the data fusion method proposed in this paper,the quantitative POI identification method and the spectral clustering method based on DTW distance are used to identify urban functional areas for POI data and cab OD data,respectively.The results show that the proposed method achieves high recognition accuracy in both overall accuracy and Kappa coefficient,which is about 7% higher than the other two methods,indicating the feasibility of the proposed method.