A New Multi-Objective Approach For Allocating Heterogeneous IoT-Based Tasks in Mobile Crowd Sensing Using Deep Learning

Abstract

Abstract Today, with the rapid growth of Internet-based service delivery services, the realization of numerous applications, including mobile mass surveillance, has become possible. In mobile mass monitoring, equipment located at the edges of the network can be used to provide computing services, storage and execution of tasks that have time priorities. Despite the many studies that have been done in the past on the application of the mobile collective monitoring approach, however, the management of heterogeneous requests considering the quality of service has not been comprehensively investigated yet. Therefore, the main goal of this thesis is to provide an approach to allocate heterogeneous tasks in the form of implementing mobile collective monitoring in such a way that both the time period for the completion of the activity is reduced and the quality of coverage and service level are observed at an optimal level. Since the participating groups in such an approach have conflict of interests, therefore, the Stackelberg inverse game theory has been used as a tool to manage the level of user participation and consider the benefit of all players. One of the features of this game model is the possibility of implementing it without having complete information of all players. In order to reach the equilibrium point of the game, the optimal strategy of the applicants is determined by using the deep reinforcement learning algorithm, because this method can be useful in finding the appropriate proposed strategy by using the history of interactions. One of the important challenges when applying learning algorithms is the lack of stability during the execution of the learning process. In this regard, an approximate policy has been used to approximate the values of the reward function, which prevents divergence during the implementation of the learning process. Another important challenge is knowing the density of user participation in mobile mass monitoring programs. In fact, the higher the number of monitoring nodes in an area, the better coverage quality can be created. For this purpose, the fuzzy system has been used, which can estimate the level of participation density by having the time range of users' presence in the study area and the level of geographic density. In this thesis, three characteristics of activity completion time frame, service quality and coverage level have been evaluated. According to the obtained results, the use of such an approach increases the coverage level by more than 17% compared to the average of common methods.