| In recent years,the number of static or mobile sensing devices in the city increases rapidly,which offers a new sensing paradigm:crowd sensing.However,crowd sensing data confronts with the common sparseness issue due to the limitation of human mobility and sensing costs,which is difficult to meet the data quality requirements of urban sensing applications represented by smart traffic.In order to ensure the effect of urban sensing applications,it is important to not only reconstruct the largely and randomly missing data but also predict the future data.However,there are two main challenges in this task:1)modeling the complex and nonlinear spatio-temporal correlation in crowd sensing data;2)solving error propagation in reconstruction and prediction.To this end,this thesis proposes a reconstruction and prediction method,which is cheap and suitable for multiple crowd sensing data,and implements a crowd sensing data quality assurance system.Firstly,this thesis proposes a multi-task learning method based on graph neural network and multi-attention mechanism,which can reconstruct and predict the crowd sensing data with multi-source contexture data.This method has the following advantages:on the one hand,it uses embedding module to extract spatio-temporal features from multi-source contexture data,and uses multi-attention module to model complex spatio-temporal correlation on graph structure;on the other hand,it alleviates the error propagation using a dynamic multi-task learning framework and a transform attention block.Extensive experiments on three real-world crowd sensing datasets demonstrate the advantages of our method over multiple state-of-the-art baselines for both data reconstruction and prediction.Finally,based on crowd sensing data reconstruction and prediction method,this thesis implements a crowd sensing data quality assurance system.It Includes two main modules of traffic flow sensing and real-time traffic prediction.The system can provide complete real-time data and multi step prediction results of crowd sensing for urban sensing applications,which helps users to understand the overall traffic status of the city and make travel plan with multiple data analysis functions. |
