基于RepVGG和LSTM两阶段移动众包任务分配算法  

Two-stage mobile crowdsourcing task assignment algorithm based on Rep VGG and LSTM

作　　者：于嵩 潘庆先 童向荣 刘庆菊 褚佳静 YU Song;PAN Qingxian;TONG Xiangrong;LIU Qingju;CHU Jiajing(College of Computer and Control Engineering,Yantai University,Yantai,Shandong 264005,China)

机构地区：[1]烟台大学计算机与控制工程学院,山东烟台264005

出　　处：《燕山大学学报》2023年第2期152-163,共12页Journal of Yanshan University

基　　金：国家自然科学基金资助项目(60903098,61502140,61572418,61472095,62072392);黑龙江自然科学基金资助项目(LH2020F023)。

摘　　要：移动众包是一种新型的感知模式,已被广泛应用于移动计算和城市生活交通服务,任务分配是移动众包中核心研究问题之一。但由于任务是动态到达的,移动众包平台在初始并不了解所有的众包任务,导致任务分配往往会陷入局部最优。为了解决上述问题,本文提出了一种基于深度学习的两阶段预测算法,第一阶段使用基于RepVGG的网络进行任务可用性的预测,第二阶段使用基于LSTM的网络进一步进行任务持续时间的预测。通过实验对比,本文所提出的算法在预测任务可用性上的准确度比传统的机器学习算法提高了32%,比同样基于深度学习的算法提高了14.2%,在预测任务持续性上的准确度相比其他算法提高了10.5%。Mobile crowdsourcing is a new type of perception model,which has been diffusely used in mobile computing and urban life transportation services.Task assignment is one of the core problems in mobile crowdsourcing.However,due to the dynamic arrival of tasks,the mobile crowdsourcing platform does not know all the crowdsourcing tasks at the beginning,resulting in the task allocation often falling into local optimum.To resolve the problems,a two-stage prediction algorithm based on deep learning is proposed.The task availability prediction is done by the Rep VGG-based network in the first stage,and in the second stage,the task duration is further predicted by the LSTM-based network.Through experimental comparison,the accuracy of the proposed algorithm in predicting task availability is 32%higher than traditional machine learning algorithms,and 14.2%higher than the algorithms based on deep learning.The accuracy of predicting task duration is 10.5%higher than other algorithms.

关 键 词：移动众包 任务分配 RepVGG LSTM 任务可用性 任务持续性 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]