Research On Key Technologies Of Sparse Mobile Crowd Sensing For Sensing Quality Assurance

In mobile crowd sensing applications,the scale of sensing tasks is large,the duration of tasks is long,and the coverage of target scenarios is wide.It is necessary to recruit enough sensing users to perform sensing tasks,resulting in high costs.Sparse mobile crowd sensing is an improvement of the traditional crowd sensing framework,which can reduce the sensing cost of sensing tasks in mobile crowd sensing.Because the data collected at different times and places have spatiotemporal correlation,the data of some sub-regions can be collected,and the data inference algorithm is used to complete the matrix to obtain a complete sensing map,thereby reducing the overall sensing cost.In sparse mobile crowd sensing,participant recruitment is particularly critical,because only a very small number of participants are selected to perform sensing tasks.The selected participants will directly affect the quality of the collected data,which in turn affects the inference of sensing data.This paper focuses on the key technologies of sparse mobile crowd sensing,mainly for user recruitment methods that affect the quality of perception.The main research work is as follows:1.A perceptual community classification method based on deep nonnegative matrix factorization is proposed.Firstly,the cluttered sensing user adjacency matrix before community classification is decomposed to obtain the mapping matrix from the original network of the sensing user to the community member space and the community member matrix of the sensing user belonging to different communities.Secondly,the mapping matrix is decomposed by deep non-negative matrix factorization,and the hidden hierarchy is learned.Finally,the encoder-decoder component is used to reconstruct the social network.Through experiments on the Gowalla dataset,the results show that the community divided by this method is closer to the real community than other comparison methods.2.A sparse sensing task allocation method based on spatio-temporal data inference is proposed.Firstly,according to the classified sensing community,the center point of the sensing community is calculated by using the semi-vector formula.Secondly,based on the different eigenvalues of the classified sensing community,the sensing platform obtains the location feature of the sensing task and matches it with the center point of the sensing community.In the final matching community,some participants are selected to complete the sensing task based on the limitation of the sensing cost.Finally,through the perceptual data provided by the participants,the complete perceptual map is obtained by using the spatio-temporal compressed sensing algorithm.Through comparative experiments with other methods,the results show that the user recruitment method proposed in this paper effectively reduces the inference error compared with other methods.3.A sensing user selection method based on long-term and short-term dynamic preferences is proposed.Firstly,the recurrent neural network model is combined with the attention network to model the user ’s short-term sequence pattern and longterm general preference at the same time,and the user ’s preference for the next stage of perception area is obtained.Secondly,the committee query algorithm is used to obtain the sub-regions allocated for the next task,and match them with the new subregions to be accessed by different users in the user set,and select the users who are most helpful for the perception map inference to perform the perception task.Finally,the complete perceptual map is obtained by data completion based on the obtained perceptual data.Simulation results show that this method accurately recommends the user ’ s next point of interest and reduces the inference error of data.