Research On Methods Of Compressive Crowdsensing Based On Matrix And Tensor Completion

Mobile crowdsensing is human-centric mobile sensing.It mainly leverages the characteristics that a person or a car always carries a large number of mobile sensors to collect information about the sensing object in the target area and uploads the data to the center so as to monitor the condition of the target area and complete the large-scale complex social sensing tasks.However,in order to obtain high-quality sensing data,a large number of participants need to be recruited to perform sensing tasks,resulting in a high sensing costs which hinders the development of mobile crowdsensing applications.Therefore,how to improve the sensing data quality and reduce the sensing cost is the core issue of mobile crowdsensing applications.To solve this problem,according to the strong temporal and spatial correlation of urban data,this paper proposes compressive crowdsensing methods based on matrix and tensor completion,which not only guarantees the sensing data quality,but also reduces the number of sensing tasks that need to be executed,thus reducing the sensing costs.This paper aims to reduce sensing costs and improve the accuracy of the sensing data.The thesis mainly focuses on three aspects,including urban data characteristics analyzing,sampling strategy and reconstruction algorithm designing.The paper proposes a new method for mobile crowdsensing data collection.The major contributions of this thesis can be summarized as follows:(1)In order to reduce sensing costs and complete high-quality data collection tasks,this paper introduces the matrix completion theory into the data collection application of large-scale mobile crowdsensing and proposes a two-stage iterative adaptive sampling strategy and reconstruction algorithm based on matrix completion.The sample cost required to reconstruct the low rank matrix data is given in the paper.This algorithm overcomes the shortcomings that the traditional uniform sampling method may miss important elements.With the adaptive sampling strategy,the reconstruction accuracy can be significantly improved for the same sample complexity.The algorithm also makes full use of the spatial correlation of the target area and the potential approximate low-rank structure of the data to reduce the number of samples that required to obtain high-quality sensing data.Simulation results demonstrate that the proposed scheme is able to not only significantly improve the reconstruction accuracy but also reduce the computation complexity comparing with the state-of-the-art matrix completion methods relying on uniform sampling.(2)To further excavate more complex high-dimensional data structures and features,this paper introduces the tensor completion technology into the application of high-dimensional data collection for large-scale mobile group intelligence perception and studies the sample cost problem and reconstruction problem of the third-order tensor data collection Based on the new tensor decomposition method-tensor singular value decomposition(t-SVD),The tubal rank and the low-dimensional subspace are used to characterize the information and structure of high-dimensional data.A two-stage iterative adaptive sampling strategy and reconstruction algorithm based on tensor completion is designed for high-dimensional data collection applications.The algorithm exploits adaptivity to identify locations which are highly informative for learning the low-dimensional subspace of the data sensor,thereby improving algorithm performance.Experimental results show that the tensor completion technique can achieve high-quality and low-cost compressive crowdsensing through sparse sampling in the region.