text
stringlengths
0
89.3k
However the second variant without kickoff achieves the
highest potential value among all variants and the benchmark
In the third variant αis set to 05 and OU noise is
deactivated for both with and without kickoff In the variant
with kickoff θkis set to 1 Testing results show similarities
to the previous two variants with shorter training cycles
However there is an increase in computational time due to
the polynomial behaviour
During the experiments we investigate the influence of OU
noise in our proposed methods It appears that OU noise couldFig 5 Bulk Good Laboratory Plant
benefit the first variant since it does not consider historical
gradient data which makes it prone to getting stuck in local
optima and facing challenges in exploring diverse states
Contrarily the impact of OU noise on the second and third
variants is minimal As a result we opt to deactivate OU noise
in both variants to simplify parameter selection
In Fig 6 a comparison is defined between the resulting
actions performance maps of player 1 the conveyor belt in
Module 1 in the BGLP The progress of the performance maps
highlights two notable learning characteristics of gradient
based learning compared to best response learning such as
a initially slower progression but then followed by faster
convergence and b a smoother exploration process guided
by the gradientbased learners
VI C ONCLUSIONS
We introduce gradientbased learning methods in SbPGs
as a replacement for the adhoc random sampling approach
used in best response learning during the training of players
policies These gradientbased learning methods are well
suited for selflearning distributed production systems We
propose three distinct variants each offering the option of
starting with or without a kickoff to establish initial weight
values Afterwards we apply these variants to a laboratory
testbed and conduct comparisons with the benchmark which is
best response learning Our testing results reveal a significant
reduction in power consumption reaching nearly 10 and
indicate that one of the variants can even enhance the potential
value compared to the benchmark Additionally the inclusion
of kickoff episodes demonstrates a substantial reduction in
training time up to 45 compared to the benchmark These
findings highlight the effectiveness and impact of gradient
based learning in SbPGs
Fig 6 A comparison of the resulted actions performance
maps of player 1 the conveyor belt in Module 1 in the BGLP
between best response learning and gradientbased learning
In future works our objective is to extend the integration of
gradientbased learning to modelbased SbPGs Additionally
we plan to explore its integration into other game structures
such as Stackelberg games
REFERENCES
1 D G Pivoto L F de Almeida R da Rosa Righi J J Rodrigues
A B Lugli and A M Alberti Cyberphysical systems architectures
for industrial internet of things applications in industry 40 A literature
review Journal of manufacturing systems vol 58 pp 176192 20212 Y N Thakare A M Karandikar V Butram A Rokade U A
Wankhade and S Honade Iotenabled environmental intelligence A
smart monitoring system in 2023 3rd International Conference on
Innovative Mechanisms for Industry Applications ICIMIA IEEE
2023 pp 3742
3 W Chen X Qiu T Cai HN Dai Z Zheng and Y Zhang Deep
reinforcement learning for internet of things A comprehensive survey
IEEE Communications Surveys Tutorials vol 23 no 3 pp 1659
1692 2021
4 M Wooldridge An introduction to multiagent systems John wiley
sons 2009
5 J Branke Multiobjective optimization Interactive and evolutionary
approaches Springer Science Business Media 2008 vol 5252
6 S Yuwono and A Schwung Modelbased learning on statebased
potential games for distributed selfoptimization of manufacturing sys
tems Journal of Manufacturing Systems vol 71 pp 474493 2023
7 D Scrimieri S M Afazov and S M Ratchev Design of a self
learning multiagent framework for the adaptation of modular production
systems The International Journal of Advanced Manufacturing Tech
nology vol 115 no 5 pp 17451761 2021
8 R S Sutton and A G Barto Reinforcement learning An introduction
MIT press 2018
9 D Schwung J N Reimann A Schwung and S X Ding Self learning
in flexible manufacturing units A reinforcement learning approach in
2018 International Conference on Intelligent Systems IS IEEE 2018
pp 3138
10 T Zhou D Tang H Zhu and L Wang Reinforcement learning with
composite rewards for production scheduling in a smart factory IEEE
Access vol 9 pp 752766 2020
11 S Lee and DH Choi Federated reinforcement learning for energy
management of multiple smart homes with distributed energy resources
IEEE Transactions on Industrial Informatics vol 18 no 1 pp 488
497 2020
12 T T Nguyen N D Nguyen and S Nahavandi Deep reinforcement
learning for multiagent systems A review of challenges solutions and
applications IEEE transactions on cybernetics vol 50 no 9 pp 3826
3839 2020
13 D Cappello and T Mylvaganam Distributed differential games for
control of multiagent systems IEEE Transactions on Control of
Network Systems vol 9 no 2 pp 635646 2021
14 J Wang Y Hong J Wang J Xu Y Tang QL Han and J Kurths
Cooperative and competitive multiagent systems From optimization
to games IEEECAA Journal of Automatica Sinica vol 9 no 5 pp
763783 2022
15 D Schwung A Schwung and S X Ding Distributed selfoptimization
of modular production units A statebased potential game approach
IEEE Transactions on Cybernetics vol 52 no 4 pp 21742185 2020