Application Research Of Graph Neural Network In Urban Computing

Urban computing takes the city as the background,integrates computing science with urban planning,transportation,energy,environment,social science and economics and other disciplines,and continuously acquires,integrates and analyzes a variety of heterogeneous big data in the city to solve the problems faced by the city.Committed to improving people’s quality of life,protecting the environment and promoting the efficiency of urban operation.Because in the process of processing city data,the original Euclidean structure can no longer fully represent some of the data structures that appear in the city.The generation of more non-Euclidean structured data need to use graph structures to display complex connections,interdependencies and influences between data.At present,the success of neural networks has promoted the research of data mining,and many machine learning tasks have been further developed.The learning method of graph data,motivated by the basic neural network model,also produces a corresponding graph neural network model.By using the graph neural network to apply neural network learning to the graph data of urban computing,we can solve the connections and dependencies between complex data.This paper mainly studies the application of graph neural network in urban computing to solve practical problems such as urban construction planning and traffic scheduling.The specific work of this paper is as follows:1、A urban land use type classification model based on graph neural network is designed and implemented.In the model,graph neural network is used to obtain the POI feature information in the plots and fuse it with the plots feature information extracted by multi-layer perception,which improve the classification accuracy.When constructing graph data,the distance relationship between plots is used to construct the weight relationship between plots,which makes it possible to aggregate node feature information according to the distance relationship between nodes during information aggregation.2、A traffic flow prediction model based on graph neural network is designed and implemented,in which the graph neural network is used to obtain the spatial feature information of traffic flow data.The graph neural network module and the time acquisition module are combined to form a spatiotemporal graph neural network for extracting the temporal and spatial feature information of the traffic flow data.In the traffic flow prediction model,the connection relationship between nodes is constructed according to the real road connection relationship,and the dual-channel information acquisition is used,so that the training process of the model can fully acquire the time and space feature information of the original data while realizing parallelization,so as to further improve the prediction accuracy.3、A demand prediction model between taxi origin and destination based on graph neural network is designed and implemented.The model uses graph neural network to obtain local feature information of original data,which solves the problem that irregular grid data cannot be processed by the convolution operation.In the demand forecasting model,the origin-destination traffic data,weather data,plot feature data and date data are used,and the feature information fusion of each data improves the prediction performance of the model.In summary,this paper mainly applies the graph neural network to solve the practical problems in urban computing,so as to achieve more accurate and realistic classification and prediction results.