Research On Mobile Users Oriented And Spatio-temporal Data Involved Crowd Sensing

With the explosive popularity of portable smart devices and the rapid development of wireless networks and communication technologies,CrowdSensing has recently become a promising paradigm for completing complex sensing and computation tasks.Combining the thought of crowdsourcing with the sensing capacity of smart devices,CrowdSensing is also an important representation in the Internet of Things,which serves the vital purpose of exploiting the ubiquitous smart devices carried by mobile users to make conscious or unconscious collaboration through the mobile networks,in order to complete large-scale and fine-grained sensing and computing tasks.Compared with traditional wireless sensor networks,CrowdSensing develops rapidly and has great advantages in deployment and maintenance,sensing range and granularity,reusability,and other aspects,which will certainly lead us into a new era of intelligence.CrowdSensing is mainly composed of two key factors,i.e.,users and data.Starting from the perspective of users and paying attention to the users' mobility,we can naturally extend CrowdSensing to Mobile CrowdSensing,that is,we can emphasize on using the widespread existence,flexible mobility and opportunity connectivity of mobile users to perform sensing tasks.Starting from the data level,by mining and utilizing the spatio-temporal correlations of the sensing data,we can extend CrowdSensing to Sparse CrowdSensing,that is,we can recruit users to sense data from a few target subareas and infer the data of the remaining unsensed subareas.Aiming at the two key factors of user and data,this dissertation studies how to make full use of users' mobility and how to effectively use spatio-temporal correlations of sensing data respectively,so as to improve the performance of CrowdSensing and reduce its costs.Furthermore,this dissertation considers the CrowdSensing on both the users' and data's sides,and studies how to use the users' mobility and the spatio-temporal correlations of sensing data at the same time.The main contributions of this dissertation are briefly summarized as follows:(1)In order to make full use of users' mobility,we propose the prediction-based user recruitment methods for Mobile CrowdSensing.First,we present a sensing-area-based mobility prediction model to obtain the probabilities that tasks would be completed by users.Based on it,we propose a greedy offline algorithm to select a set of users under a budget constraint,and a better approximate ratio is achieved.Furthermore,we extend the user recruitment problem to a more realistic online setting where users come in real time and propose an online algorithm,which also achieves a good competitive ratio.Finally,we design a distributed user recruitment framework Crowd UserS and implement an Android prototype system.(2)In order to make effective use of spatio-temporal correlations of sensing data,we propose the subarea selection methods for enhancing data inference accuracy in Sparse CrowdSensing.First,we propose a compressive sensing-based data inference algorithm with spatio-temporal constraints.Based on this,this dissertation first studies the subarea selection problem in a single-task scenario and proposes a reinforcement learning-based subarea selection algorithm for Sparse CrowdSensing.Furthermore,the single-task scenario is extended to the multi-task scenario,and a multi-dimensional subarea selection algorithm is proposed in Sparse CrowdSensing.In addition,a two-stage online training framework is proposed to reduce the dependence of the proposed algorithms on a large amount of training data.(3)In order to make use of both users' mobility and spatio-temporal correlations of sensing data,we propose a user recruitment strategy for Sparse Mobile CrowdSensing,based on the mobility prediction and data inference.Considering the variable user mobility and complicated data inference,we propose a three-step user recruitment strategy from the user's and data's sides,including user selection,subarea selection and user-subarea-cross selection.We first select some candidate user sets,which may cover the most subareas under the budget constraint(user selection),then estimate which subareas are more useful on data inference according to the selected candidates(subarea selection),which finally guides us to recruit the best user set(user-subarea-cross selection).In conclusion,this dissertation proposes a series of models and methods based on mobility prediction and data inference,in order to make full use of users' mobility and spatio-temporal data correlations in CrowdSensing.The theoretical analysises and extensive experiments verify the effeciveness of the proposed models and methods,which can be used as theoretical and technical references for CrowdSensing applications in different domains.