An incentive mechanism design for mobile crowdsensing with demand uncertainties
Introduction
With the ubiquity of mobile devices such as smartphones and tablets that are equipped with multiple powerful built-in sensors including GPS, accelerometer, gyroscope, camera, etc., the mobile crowdsensing (MCS) applications which provide location based services [1] become possible. Currently, various of MCS systems [2], [3], [4], [5] have been deployed that cover almost every aspect of our lives, including healthcare, intelligent transportation, environmental monitoring, etc.
In the MCS system that offers crowdsensing applications, the sensing-platform (SP) will recruit mobile users (MUs) at locations of interest to report sensing data. Many of existing MCS systems [6], [7] are based on the voluntary participation from MUs. However, to perform the sensing tasks, the participating MUs have to consume their own resources such as computing and communicating energy. Moreover, the MUs may face the potential privacy threats when the sensing data is submitted with own sensitive information (e.g. location tags and visiting patterns). For these reasons, the MUs would not be interested in participating in the sensing tasks unless they receive a satisfying reward to compensate their resources consumption and potential privacy breach. Therefore, it is necessary to design an effective incentive mechanism that can stimulate the MUs to participate in the crowdsensing applications. In order to achieve the maximum user participation level, large quantities of incentive-aware mechanisms [8], [9], [10], [11], [12] have been proposed by research community for the MCS systems. Notably, in real practice, the smart devices’ resources such as energy are limited, and these resources need to satisfy MUs’ varying demand caused by their uncertain behavior (e.g., when MUs are busy at work, their smart devices may be free. When MUs want to have entertainments, their smart devices may be occupied with few resources left). However, few of these aforementioned works take MUs uncertain behavior into consideration. Therefore, the design of incentive mechanism for MCS game with demand uncertainties is still an open problem.
To deal with this problem, in this paper, the interaction between SP and MUs is formulated into a two-stage Stackelberg game. As shown in Fig. 1, in Stage I, the SP as the leader of the Stackelberg game first determines and broadcasts its pricing policy. In Stage II, each MU as a follower computes his or her sensing effort based on the price offered by the SP, his or her resources constraints and demand uncertainties. The analysis in this two-stage problem is particularly challenging, as we need to characterize the SP’s profit by first computing the MUs’ sensing effort with demand uncertainties. Through mathematical analysis, the existence and uniqueness of the Stackelberg Equilibrium (SE) in this MCS game is proven and the expressions for computing the SE is derived. That is, the SP in Stage I has an optimal pricing strategy and the MUs in Stage II also have optimal decisions under their own demand uncertainties.
However, in order to compute the SE of the above static MCS game, the SP needs to know the private information of the MUs, which is impossible in lots of practical situations. To protect MUs’ private information, the dynamic MCS game is modeled and dynamic incentive mechanism based on deep reinforcement learning (DRL) approaches are employed, which enable the sensing platform to learn the optimal pricing strategy directly from game experience (the past game records). Since the game experience of the SP can be regarded as a motivation for its future pricing strategy, the dynamic MCS game can be formulated into a Markov Decision Process (MDP) problem. Thus, it can be addressed by DRL algorithms effectively [13].
Overall, the main contributions of this paper can be summarized as follows:
- 1.
A novel economic model for the MCS game with MUs’ resources constraints and demand uncertainties is formulated and an incentive mechanism based on a two-stage Stackelberg game is designed.
- 2.
The existence and uniqueness of the SE in the proposed MCS game is proven and its computing procedure is provided, revealing the feasibility of allowing MCS game to cope with MUs’ uncertain demand and limited resources.
- 3.
A dynamic incentive mechanism (DIM) based on DRL approach for the dynamic MCS game is proposed, which enables the SP to learn the optimal pricing strategy directly from game experience without any prior knowledge about MUs’ private information.
- 4.
Numerical simulation results demonstrate the effectiveness of the proposed incentive mechanisms for both of the static MCS game and the dynamic MCS game. It is also derived that the demand uncertainties have a significant impact on MCS system performance.
The rest of the paper is organized as follows. Section 2 provides a literature review. Section 3 presents the network economics model of the crowdsensing system. The incentive mechanism based on a two-stage Stackelberg game for the static MCS game is designed in Section 4 and the DRL-based dynamic incentive mechanism for the dynamic MCS game is designed in Section 5. In Section 6, the numerical simulations are conducted to evaluate the performance of the proposed incentive mechanisms, followed by conclusions of this paper in Section 8.
Section snippets
Literature review
MCS has been widely studied in recent years [1]. For example, Reddy et al. [14] developed an application to enable sensing platform employ well-suited participants to complete sensing tasks. Xiao et al. [6] and Li et al. [15] both studied the task allocation and participants selection problem in MCS. However, these works only focus on the user selection, task assignment or sensing data collection. They do not consider the design of incentive mechanism, which has been widely studied in lots of
System model and problem formulation
We consider a single SP which resides in the cloud and consists of some servers. A set of MUs that connect to the sensing-platform via the Internet. The sensing-platform will stimulate the mobile users to participate in the MCS tasks via rewards. More specifically, the system model is described in Section 3.1.1, followed by the problem formulation in Section 3.2 finally.
Incentive mechanism for static MCS game
In this section, how to design the incentive mechanism for the static MCS game by solving the Stackelberg game defined in Section 3.2 is demonstrated. In the static MCS game, the main challenges are (a) how to develop the resource allocation strategy for the MUs and (b) how to develop a pricing strategy for the SP. In the following, it is firstly proven that for any feasible each MU has a unique optimal resource allocation strategy in the second stage (Section 4.1). Afterwards,
Dynamic incentive mechanism for MCS with deep reinforcement learning approach
In this section, a dynamic incentive mechanism (DIM) based on deep reinforcement learning (DRL) approach is designed for MCS. Obviously, in order to derive the optimal pricing strategy, the SP needs to solve Eq. (9) by knowing the private information such as τn, δn, cn of the MUs. However, it is unrealistic to obtain these information in practice because MUs may refuse to expose their utility parameters due to privacy concerns. Hence, a DRL approach is designed to learn the optimal strategy
Performance evaluation
In this section, numerical simulations are conducted. Specifically, 5 MUs are randomly generated. For each MU, cn and δn are randomly from [0,1] while guaranteeing δn > cn. We set the total available resources τn of each MU to 20 units, and randomly drawn the own resources demand ξn from a uniform distribution in [0,25].
Improvement of DIM
Although the proposed DIM can effectively learn the SE in MCS game, it requires large quantities of interactions between SP and MUs, leading to relatively high time cost to reach the optimal pricing strategy. Moreover, in practice, the MCS system may not allow the SP to interact constantly with MUs and limit the number of learning episodes. Therefore, it is significant to improve the learning efficiency of DIM, especially when the training data is limited. One promising direction is to leverage
Conclusion
In this paper, the static MCS game with MUs’ resources constraints and demand uncertainties is formulated firstly, the incentive mechanism is then considered based on a Stackelberg game. The existence of the unique SE is proved and the expressions for calculating the SE are provided. By analyzing the SE, it is found that the MUs’ demand uncertainties have evident impacts on the performance of the MCS system. Moreover, considering that the SP requires the MUs’ private information to achieve the
CRediT authorship contribution statement
Yufeng Zhan: Writing - original draft. Yuanqing Xia: Conceptualization. Jiang Zhang: Investigation. Ting Li: Validation. Yu Wang: Conceptualization.
Declaration of Competing Interest
None.
Acknowledgment
This work was supported by National Natural Science Foundation under Grant 61836001, the National Natural Science Foundation Projects of International Cooperation and Exchanges under Grant 61720106010, the Foundation for Innovative Research Groups of the National Natural Science Foundation of China under Grant 61621063, the National Natural Science Foundation of China 61572347, the US National Science Foundation (CNS-1319915 and CNS-134335), and the U.S. Department of Transportation Center for
References (42)
- et al.
Online task assignment for crowdsensing in predictable mobile social networks
IEEE Trans. Mob. Comput.
(2016) - et al.
Distributed algorithms to compute Walrasian equilibrium in mobile crowdsensing
IEEE Trans. Ind. Electron.
(2016) - et al.
Spectrum trading in cognitive radio networks: acontract-theoretic modeling approach
IEEE J. Sel. Areas Commun.
(2011) - et al.
A Course in Game Theory
(1994) - et al.
Mobile crowd sensing and computing: The review of an emerging human-powered sensing paradigm
ACM Computing Surveys
(2015) - et al.
Nericell: rich monitoring of road and traffic conditions using mobile smartphones
Proc. of ACM SenSys
(2008) - et al.
VTrack: accurate, energy-aware road traffic delay estimation using mobile phones
Proc. of ACM SenSys
(2009) - et al.
AirCloud: a cloud-based air-quality monitoring system for everyone
Proc. of ACM SenSys
(2014) - et al.
SRMCS: a semantic-aware recommendation framework for mobile crowd sensing
Inf. Sci.
(2018) - et al.
Ear-phone: an end-to-end participatory urban noise mapping system
Proc. of ACM IPSN
(2010)
Incentive mechanism in platform-centric mobile crowdsensing: a one-to-many bargaining approach
Comput. Netw.
An exchange market approach to mobile crowdsensing: pricing, task allocation, and Walrasian equilibrium
IEEE J. Sel. Areas Commun.
Incentive mechanisms for crowdsensing: crowdsourcing with smartphones
IEEE/ACM Trans. Netw.
Free market of crowdsourcing: incentive mechanism design for mobile sensing
IEEE Trans. Parallel Distrib. Syst.
Trust region policy optimization
Proc. of ICML
Examining micro-payments for participatory sensing data collections
Proc. of ACM UbiComp
Dynamic participant recruitment of mobile crowd sensing for heterogeneous sensing tasks
Proc. of IEEE MASS
Map: multiauctioneer progressive auction for dynamic spectrum access
IEEE Trans. Mob. Comput.
Self-interest-driven incentives for ad dissemination in autonomous mobile social networks
Proc. of IEEE INFOCOM
OURS: optimal unicast routing systems in non-cooperative wireless networks
Proc. of ACM MobiCom
Conflicts and incentives in wireless cooperative relaying: a distributed market pricing framework
IEEE Trans. Parallel Distrib. Syst.
Cited by (29)
SCTD: A spatiotemporal correlation truth discovery scheme for security management of data platform
2023, Future Generation Computer SystemsCitation Excerpt :The second is how to identify the trust of workers [6,13,17]. The third is how to recruit trustworthy workers while minimizing system costs, thereby minimizing the risks and potential attacks suffered by the data platform [9,15]. The first challenge in identifying the truth of the reported data is the most critical issue.
Privacy preserving IoT-based crowd-sensing network with comparable homomorphic encryption and its application in combating COVID19
2022, Internet of Things (Netherlands)Crowdsensing based missing data inference algorithm considering outlier data and GPS errors
2022, Information SciencesCitation Excerpt :A mobile crowdsensing system (MCS) is a system that uses mobile devices to effectively collect data, so as to provide a variety of novel Internet of Things applications [18], as shown in Fig. 1. An MCS needs to recruit a large number of participants by utilizing efficient incentive mechanisms [35]. The mobile devices of participants automatically collect the data required by the MCS and then transmit these data to the server of the MCS through various communication methods, such as 4G, 5G, or WiFi.