Research On Incentive Mechanisms And Quality Aware Data Collection For Crowd Sensing Systems

In recent years, due to the rapid development of communications, inte-grated circuits, software and microelectronics, the mobile terminal equipment that previously had only communication functions, has become more and more powerful in sensing, computing and storage. Because of the development of smart phones and mobile communications networks, the original common com-munication tools have become the powerful function of intelligent communi-cation terminals. At the same time, the emergence of new smart devices, e.g.,tablet PCs, wearable devices and even unmanned aerial vehicles, further ex-pands the scope of smart devices. Based on this context, with the popularity of these intelligent terminals, the smart devices already have the ability to re-place traditional sensors to collect data, which prompted researchers to propose a new way to collect data, i.e., Crowd Sensing. Different with traditional sensor networks, the crowd sensing campaigns do not require to install sensors before collecting data. The user with the intelligent terminal acts as the basic sensing unit. The users employ their own smart devices to collect and upload (share)sensing data.In crowd sensing systems, people use their smart devices to collect sur-rounding environmental data. The ordinary people are both "the consumers and the beneficiaries" of the collected sensing data, and the "producers" of the sensing data. The crowd sensing system needs to have ordinary people as par-ticipants to join the campaigns. Different from the traditional sensors, people have their own wishes and habits. Each person’s activity time and route often have a great probability of uncertainty and randomness, resulting in sensing data cannot be collected in time or reach the required quantity. On the other hand, the performance of sensors built in various type of smart devices is dif-ferent, and people upload false sensing data. All of these conditions will lead to differences in the accuracy, completeness and timeliness of the collected data,and cannot reflect the real situation of the sensing area. Finally the sensing re-sult cannot satisfy the requirement of the platform and task publishers. The less quantity and low quality of sensing data has inevitably become a critical prob-lem that constrains the development of crowd sensing campaigns. Because of this, how to select the proper participants to provide high quality sensing data,how to encourage participants to provide sensing data are the key challenges that need to be solved urgently. Although there are a variety of crowd sensing applications, there is still a lack of effective data collection methods and reason-able incentive mechanisms. Despite it has accumulated more research results to solve the data collection problem in traditional sensor network, these results are hardly to directly meet the crowd sensing requirements, and they almost do not consider the incentive mechanisms. In summary, in this paper, the main contributions of the thesis are as follows:First, since participating crowd sensing systems consumes resources from the participants’ smart devices (e.g., electricity, network bandwidth usage), a mechanism is needed to compensate participants for the cost of collecting data.Different from the traditional sensor network, participants in crowd sensing sys-tems have great autonomy and subjectivity. Data is the fundamental of crowd sensing campaigns. The data comes from the participant’s active collection activities. Therefore, the incentive mechanism is an indispensable part of the crowd sensing systems, who can encourage participants to provide more quan-tity and higher quality of sensing data. Therefore, in this thesis, besides con-sidering to collect maximum amount of sensing data in order to meet the re-quirement of the platform, the mechanism also considers maintaining a certain number of participants. Imagine if a task costs a participant’s device too much energy, but the final return does not compensate for the cost of collecting data,then she may take a negative attitude towards the subsequent sensing tasks.Therefore, we design a difficult of task (DoT) index to measure the difficulty of each task, allowing participants to select sensing task according to the task difficulty index under their unit reward. In addition, in order to ensure that more and more participants perform the sensing task, a participant is allowed to perform only one task in a period,the remaining tasks need to be performed by other unselected participants.Second, with the development of science and technology, the types of smart devices that can sense and collect data increase. For example, vehicles equipped with on-board units (OBUs) have the ability to sense and share their surrounding environment. The new sensing scenario has brought new prob-lems. In the past, many incentive mechanisms focus on the offline scenario,which requires many participants to exist in the sensing area at the same time.This scenario is adopted when participants move slowly and they may have been in the sensing area during the execution of the sensing task. However, if the sensing device is the vehicle with fast speed, then the offline scenario is no longer apply the incentive mechanism. In this case, a new incentive is needed that immediately decides whether to select the participant based on the partic-ipant’s own attributes (e.g., requested reward, reputation value.). For this kind of online scene, we propose an online incentive mechanism. The mechanism takes into account both the platform and participant’s benefits. For the plat-form, the proposed incentive mechanism selects credible participants to collect data as much as possible. For the participant, besides her requested reward, the participant may also receive additional reward based on her reputation value in order to encourage her to provide more sensing data. The proposed incentive mechanism selects participants as many as possible under the budget constraint,in order to let more participants can be rewarded.Last, sensing data collection is always one of the important issues in the field of crowd sensing research. One of the problems with sensing data col-lection is the quality of collected data. If there is not enough amount of high quality sensing data, then the task publisher may no longer publish the task,without task the platform does not need to select participants to perform, and ultimately lead to the entire crowd sensing system lockout. In order to solve the above problem, this thesis proposes a single task-based data quality prediction method for the GPS data in the road traffic environment monitoring system.The method is designed based on the real scenario, that is, the participants ar-rive at different time in random order. When a participant arrives, the proposed method first uses a prior quality method to predict the amount of high quality data she may upload, then decides whether to select her according to the pre-dicted results and her requested reward. Different from the reputation definition and update method, the proposed method uses a binomial-poisson distribution(BPD) to model the amount of high quality sensing data that the participant may upload. The expectation maximization method (EM method) is used to estimate the parameter values in the binomial-poisson distribution.Finally, the biggest feature of crowd sensing campaigns is that it does not only use professional sensors, but also uses the participant’s mobile smart de-vices to collect data. Despite crowd sensing systems can save deployment over-head, because the data collection campaign may interfere with other activities of the participant (e.g., calling, sending text messages, daily work and entertain-ment), and requires the participant’s permission before collecting data. More-over, data provided by different participants in terms of accuracy and timeliness is different, which means that some data cannot be a good response to the real situation of the sensing area. In order to solve the above problem, we pro-pose a mechanism to maximize the data credibility under the budget constraint.Here the overall data credibility is influenced by two aspects: the credibility of participants and their position distribution. Participants with high reputation values are more likely to provide credible sensing data. At the same time, data collected from different locations can fully represent the overall condition of sensing area. Therefore, in this thesis, we divide the maximizing data credi-bility problem into two sub-problems: one is to select credible participants to maximize the total credibility values, the other is to select participants from dif-ferent sensing blocks to maximize the sensed area. These two sub-problems are combined into a multi-objective optimization problem, and a credible partici-pant selection method is designed to solve it.The simulation results obtained by using the real data set show that the proposed methods in this thesis can better solve the problems emerged in data collection and incentive mechanism, and satisfy both the platform and partici-pants’ benefits, compared with the existing research.