Fine-Grained Traffic Flow Inference With Missing Values For Smart Cities

In recent years,with the vigorous development of artificial intelligence technology,the construction of smart cities has gradually become one of the important directions of urban development,and its application fields have covered many aspects such as urban management,life,and economy.In a smart city,traffic flow is a very important indicator,which involves many aspects such as urban traffic operation,traffic congestion,and urban planning.At present,most studies focus on the change of urban traffic flow,and have achieved rich research results,but there are still some deficiencies in the research on fine-grained flow inference with missing values.This paper aims to conduct in-depth research on two aspects of urban traffic flow forecasting and flow inference with missing values.Urban regional traffic flow forecasting plays an important role in the construction of smart cities.However,the existing research on urban regional traffic flow forecasting mainly focuses on predicting the future traffic flow of the entire urban area based on the complete flow map at historical moments.The research problem of this paper is the work of predicting the complete flow map at the current time based on the incomplete flow at the historical time.First of all,the specific definition and related symbolic description of the problem is given,and then the urban flow prediction model of time and spatio-temporal learning is proposed,which not only designs a spatio-temporal attention module for the time-continuous attribute of the flow graph and the adjacent attribute of the spatial region,Moreover,two spatiotemporal attraction constraint loss functions are proposed for spatial rotation and temporal periodicity to constrain the network to further improve the spatiotemporal feature extraction ability of the model.All the experimental results demonstrate the effectiveness of the proposed model,which has the advantages of few model parameters,low hyperparameter sensitivity and no dependence on external datasets.The urban fine-grained flow inference task is an important issue in urban regional flow forecasting,and its research provides great help for reducing the number of sensors that need to be deployed in smart cities,but the current problem in real scenarios is the lack of original coarse-grained traffic flow maps.While addressing these problems,this paper also designs a general framework suitable for spatio-temporal data mining,considering the complex relationship between coarse-grained and fine-grained traffic.model of the encoder-decoder architecture,and the other is a fine-grained decoder that only incorporates spatial attention learning.While solving the problem of coarse-grained missing value prediction,it can further decode coarse-grained spatio-temporal features,and carry out reconstruction and dimensionality reduction of fine-grained features.A large number of experimental results prove that the proposed method is superior in the prediction of coarse-grained missing value flow,and it is also advanced in fine-grained flow inference with missing values.The first work of this paper solves the problem of incomplete data prediction in urban flow forecasting,and on this basis extends the work of fine-grained flow inference with missing values based on multi-task learning.At the algorithm level,the algorithms of the two tasks have continuity,consistency,and scalability.