A deep reinforcement learning based research for optimal offloading decision DOI Creative Commons
Jianji Ren, Donghao Yang, Yongliang Yuan

et al.

AIP Advances, Journal Year: 2023, Volume and Issue: 13(8)

Published: Aug. 1, 2023

Currently, a concern about power resource constraints in the distribution environment is being voiced increasingly, where increase of consumption devices overwhelms terminal load unaffordable and quality cannot be guaranteed. How to acquire optimal offloading decision resources has become problem that needs addressed urgently. To tackle this challenge, novel reinforcement learning algorithm named Deep Q Network with partial strategy (DQNP) proposed optimize allocation for high computational demands. In DQNP, coupled coordination degree model Lyapunov are introduced, which trade-offs decouples relationships between local-edge latency–energy consumption. derive decision, computation utility function selected as objective function. addition, pruning availed further improve training time inference results. Results show mechanism can significantly decrease value decline weighted sum latency energy by an average 3.61%–7.31% relative other state-of-the-art algorithms. Additionally, loss process successfully mitigated; furthermore, effectiveness also verified.

Language: Английский

Evaluation of new sparrow search algorithms with sequential fusion of improvement strategies DOI
Jun Li, Jiumei Chen, Jing Shi

et al.

Computers & Industrial Engineering, Journal Year: 2023, Volume and Issue: 182, P. 109425 - 109425

Published: July 7, 2023

Language: Английский

Citations

16
An Improved Black Widow Optimization Algorithm for Engineering Constrained Optimization Problems DOI Creative Commons
Dongxing Xu, Jianchuan Yin

IEEE Access, Journal Year: 2023, Volume and Issue: 11, P. 32476 - 32495

Published: Jan. 1, 2023

In solving engineering constrained optimization problems, the conventional black widow algorithm (BWOA) has some shortcomings such as insufficient robustness and slow convergence speed. Therefore, an improved (IBWOA) is proposed by combining methods of double chaotic map, Cauchy center gravity inverse difference mutation golden sine guidance strategy. Firstly, quality initial population BWOA based on map; Secondly, in order to make full use information between current optimal position thus improve accuracy, (Gold-SA) introduced update individuals; Finally, barycenter reverse differential operator employed increase diversity population, avoid local global search ability algorithm. addition, characteristics IBWOA are analyzed Markov process probability reaches 1 for globally solution. The performance was evaluated eight continuous / discrete hybrid problems typical benchmark functions. results show that can speed effectively comparing with other algorithms.

Language: Английский

Citations

14
A deep reinforcement learning based research for optimal offloading decision DOI Creative Commons
Jianji Ren, Donghao Yang, Yongliang Yuan

et al.

AIP Advances, Journal Year: 2023, Volume and Issue: 13(8)

Published: Aug. 1, 2023

Currently, a concern about power resource constraints in the distribution environment is being voiced increasingly, where increase of consumption devices overwhelms terminal load unaffordable and quality cannot be guaranteed. How to acquire optimal offloading decision resources has become problem that needs addressed urgently. To tackle this challenge, novel reinforcement learning algorithm named Deep Q Network with partial strategy (DQNP) proposed optimize allocation for high computational demands. In DQNP, coupled coordination degree model Lyapunov are introduced, which trade-offs decouples relationships between local-edge latency–energy consumption. derive decision, computation utility function selected as objective function. addition, pruning availed further improve training time inference results. Results show mechanism can significantly decrease value decline weighted sum latency energy by an average 3.61%–7.31% relative other state-of-the-art algorithms. Additionally, loss process successfully mitigated; furthermore, effectiveness also verified.

Language: Английский

Citations

0