Functional Identification Of Urban Land Space Based On Multi-source POI And Machine Learning

In the accelerating urbanization process in China,urban land functions tend to lose balance in development,which in turn affects the utilization of urban land resources and the quality of urban life.The state has repeatedly emphasized the importance of sustainable urban development capacity in policy promulgation and in the implementation of the territorial spatial planning system.Through the identification of urban functional areas to understand the structure of urban functional areas can provide better decision support for urban spatial planning and development construction,and the classification of urban functional areas is a prerequisite for achieving scientific and reasonable planning and construction,which helps the rational use of urban spatial resources,the optimization of urban structure,and the scientific and healthy development of cities.At present,scholars at home and abroad have achieved many breakthrough results in identifying urban territorial spatial functions by constantly changing perspectives and methods,but the data sources of existing research results are mostly based on remote sensing and some platforms need to purchase to obtain information,which has the problems of difficult data acquisition,low timeliness,expensive and data acquisition cannot directly realize some function identification.The methods about the research of urban land space function identification of multi-source geographic big data are still being drilled,so this paper is based on the advantages of POI data,through the integration of multi-source POI data and machine learning algorithm to achieve urban land space function identification and division,using Nanning City 2019 Gaode POI data,Baidu POI data for spatial analysis,comprehensive 2019 OSM road network data,and take the central urban area of Nanning City(with the Outer Ring Road as the boundary)as the study area to identify and classify urban land spatial functions.(1)A method of multi-source POI data fusion based on multi-feature similarity is proposed to explore the automated fusion method of multi-source multi-feature POI to provide high-quality data support for scientific identification of urban functional area types.For the fusion of multi-source heterogeneous POI data on the basis of comparing three feature similarities of name,spatial location distance and address information by adding category information features and optimizing the calculation of address information feature similarity through the word separation method,the POI similarity matching is scientifically calculated using the hierarchical analysis method.Experimentally proposed method based on weighted multi-attribute similarity of multisource heterogeneous POI matching method comparison,the results show that this paper proposed multi-source POI matching method comparison to be better,more suitable for multi-source heterogeneous POI data after simple data pre-processing direct matching,with a certain data accuracy can be used for multi-source POI data matching.(2)A method of urban territorial spatial function identification by fusing multisource POI and machine learning is proposed to provide intelligent decision support for urban territorial spatial planning.Based on the effective fusion of multi-source POI data,the KStar algorithm with kernel density analysis and machine learning is used to intelligently distinguish and accurately identify the urban territorial spatial functions in Nanning city.The results of the study show that it is feasible to combine machine learning to identify spatial functions in urban areas of Nanning,and the use of machine learning reduces the interference of human subjective consciousness to a certain extent and makes the results more scientific.The use of fused multi-source POI data improves the quality of data compared with single-source POI data and achieves optimization of data sources,which helps to improve the recognition accuracy.After random sample verification,the accuracy reaches 86.46%,which is also a breakthrough in recognition accuracy compared with previous studies.The machine learning KStar algorithm proposed in this paper has higher recognition accuracy and better method than the weighted quantitative recognition method.The data source is relatively easy and fast to obtain,and the machine learning reduces the interference brought by human subjective consciousness and improves the intelligence and automation,which is a good method to explore the spatial function classification of urban land.