獎勵回溯DQN驅(qū)動的多QoS工業(yè)網(wǎng)絡(luò)時隙調(diào)度方法
打開文本圖片集
中圖分類號：TP393 文獻標志碼：A 文章編號：1001-3695（2025）07-029-2141-06
doi： 10.19734/j. issn.1001-3695.2024.12.0491
Abstract：Existing researchonmulti-QoSscheduling problems，due toitsreliancesolelyonimmediatereward feedback mechanisms，faces isues ofpoor scalabilityand resource wastagewhen handlingdelay-sensitivedataand mediadata withcontinuous transmision requirements inresource-constrained scenarios.To addressthis problem，this paper proposed aRB-DQN algorithm.Thisalgorithmadjustedthecurrntstate’spolicyevaluationbybacktrackingfutureinteractions，effectivelyidentifyingandresolving packetlosscausedbysuboptimalschedulingstrategies.Additionaly，itdesignedaLTTmetric，whichcomprehensivelyconsideredtheservicerequirements ofbothdelay-sensitivedataandmedia-typedata，alowing forweightadjustmentstoemphasizediferentpriorities.Extensivesimulationresultsdemonstratethattheproposedalgorithmsignificantlyreducesthe delayand jiterofdelay-sensitivedata while ensuringthe smothnessandstabilityof media-type data，outperforming other scheduling strategies.
Keywords：time slot scheduling；deep reinforcement learning；multi-QoS；reward backtracking
0 引言
隨著工業(yè)互聯(lián)網(wǎng)的快速發(fā)展，制造業(yè)正經(jīng)歷深刻的變革。（剩余15719字）