Research On Data Quality Aware Task Allocation And Incentive Mechanisms In Mobile Crowdsensing

The conception of the Internet of Things(IoT)refers to the integration of sensors and common embedded computing devices on the edge of the Internet,and achieving comprehensive and large-scale sensing and connection of various physical objects and their surrounding objects.The devices connected to the Internet of Things are various and the used communication technologies are different,and we can use the intelligent computing technology such as data mining to analyze the large amount of sensing data collected by the sensor network to obtain useful information for human decision-making and behavior guidance.With the popularity of mobile devices such as smart phones and smart hand rings,human-centered mobile sensing networks are bringing about changes for the Internet of Things.Mobile crowd sensing system combines the advantages of concepts of crowdsourcing systems and mobile sensing networks.It makes use of a large number of mobile sensing devices held by ordinary citizens and existing data transmission facilities(such as 4/5G,WIFI base station,etc.)to form a large-scale,fine-grained and complex sensing network.It not only expands the coverage areas of existing sensor networks,but also reduces the cost of infrastructure maintenance.It has been widely used for providing large-scale and complex sensing services such as environmental monitoring,public services,traffic forecasting and so on.With the in-depth study of the application and theory of mobile crowd sensing,its problems in quality control,resource optimization,participants incentives and privacy protection have become increasingly prominent.The key issues need to be solved urgently are those problems such as formulating reasonable data collection and quality assurance mechanism,balancing and efficient distribution of mobile participants' sensing resources,appropriate incentive mechanism and privacy protection strategy to ensure peoples participating actively and effectively.In view of the above problems,this paper fully studies existing mobile crowd sensing systems,and proposes a general measurement model of the participants' reliability and data quality,and further gives the concept of quality coverage according to this model.In view of the different scenarios of single task allocation and multiple task allocation,this paper proposes the optimization problem of the trade off between incentive cost and coverage quality.Finally we use simulation data and real data sets to evaluate the high performance of our proposed approximation algorithms.In single task allocation,considering the different scales of sensing regions,we propose the optimization problems of maximizing the minimum source spatial quality coverage and maximizing the total spatial quality coverage under the constraint of incentive cost.Considering the NP-hardness of these two optimization problems,we propose a heuristic algorithm and an approximate greedy algorithm to solve these two problems approximately.By simulating different sensing scenarios with simulated data and real data sets,we compare our algorithm with existing algorithms in terms of quality assurance,cost control and participant incentives.It proves that under the moderate budget control,comparing to k-coverage algorithm our algorithms can improve the average quality coverage by about 15% to 35%,and comparing to the existing incentive strategy of unified unit bonus reward,they can improve the task completion coverage ratio by about 0.1.In multiple task allocation,we consider the overlap of data types and spatial-temporal regions of tasks submitted by different data requesters.We present a measurement model for the quality of tasks in multiple task scenarios,and propose an optimization problem to minimize incentive costs while ensuring the quality of each task in each allocation cycle.In order to ensure the computational efficiency,we propose a greedy strategy that prioritizes matching participants and tasks with maximum quality contribution per unit cost.Finally,simulation experiments on real data sets prove that our algorithm can save 5% to 15% and 25%-45% incentive costs respectively compared with the maximum quality coverage first algorithm and the minimum cost first algorithm under the constraint that the quality coverage level of each task meets the given threshold.