Time-Constrained Task Allocation and Worker Routing in Mobile Crowd-Sensing Using a Decomposition Technique and Deep Q-Learning

Abstract

Mobile crowd-sensing (MCS) is a data collection paradigm, which recruits mobile users with smart devices to perform sensing tasks on a city-wide scale. In MCS, a key challenge is task allocation, especially when MCS applications are time-sensitive, and the platform needs to consider task completion order (since a worker may perform multiple tasks and different task completion orders lead to different travel costs and response times, i.e., the times needed to arrive at the task venues), requirements of tasks (such as deadline and required sensor) and workers heterogeneity. In other words, the task allocation problem consists of multiple task completion order problems, which is challenging to solve due to the large solution space. Therefore, in this paper, we first formulate the considered problem into two related integer linear programming problems (i.e., assignment and task completion order problems) using a decomposition technique in order to reduce the problem size and enable the use of diverse searching strategies. Then, a deep Q-learning (DQN)-based algorithm, namely assignment DQN with a local search (A-DQN w/ LS), is proposed to determine the task-worker assignments, which iteratively employs an asymmetric traveling salesman (ATSP) heuristic to find the task completion orders of the workers. The local optimizer is applied at the end of the A-DQN algorithm to deal with the computation time and local optima. Simulation results show that the proposed method outperforms existing approaches under different sensing dynamics in terms of total cost. INDEX TERMS Deep reinforcement learning, mobile crowd-sensing, task allocation, tabu search

Matrix Factorization-Based Trajectory Completion for Addressing Missing Trajectories in Mobile Crowd Sensing

“…Matrix Completion (MC) algorithms have been widely applied in various fields, including image processing, network traffic handling, recommendation systems, and more [14][15][16][17][18][19].…”

Dynamic Task Assignment Framework for Mobile Crowdsensing with Deep Reinforcement Learning

“…Ji et al [32] construct a dynamic task allocation model and proposes a Q-learning-based hyperheuristic evolutionary algorithm to maximize the average perceived quality of all tasks in each period. Akter et al [33] proposed a deep Q-learning-based algorithm to determine the assignment of tasks and workers and iteratively used the asymmetric travelling salesman (ATSP) heuristic to find the task completion order of workers. Wang et al [34] proposed a privacy-enhanced multiregional task assignment strategy (PMTA) for Healthcare 4.0 using deep differential privacy, deep Q network, federated learning, and blockchain to effectively protect the privacy of tasks and patients and obtain better system performance.…”

Task Allocation Mechanism Based on Ant Colony Algorithm in Distributed System

“…Therefore, in the process of ant colony solution construction, pheromones in the global optimal solution and the optimal solution of the current iteration need to be updated in order to avoid the algorithm falling into the local optimal solution due to too fast convergence speed. These characteristics are an embodiment of the learning function of ACA [18]. In ACA, ρ is generally used to reflect the rate of pheromone volatilization, the size of the pheromone volatilization rate ρ directly affects the convergence and global search ability of ACA.…”