TY - GEN
T1 - A Multi-agent Deep Reinforcement Learning-Based Collaborative Willingness Network for Automobile Maintenance Service
AU - Hao, Shengang
AU - Zheng, Jun
AU - Yang, Jie
AU - Ni, Ziwei
AU - Zhang, Quanxin
AU - Zhang, Li
N1 - Publisher Copyright:
© 2022, The Author(s), under exclusive license to Springer Nature Switzerland AG.
PY - 2022
Y1 - 2022
N2 - With the growth of maintenance market scale of automobile manufacturing enterprises, simple information technology is not enough to solve the problem of uneven resource allocation and low customer satisfaction in maintenance chain services. To solve this problem, this paper abstracts the automotive maintenance collaborative service into a multi-agent collaborative model based on the decentralized partially observable Markov decision progress (Dec-POMDP). Based on this model, a multi-agent deep reinforcement learning algorithm based on collaborative willingness network (CWN-MADRL) is presented. The algorithm uses a value decomposition based MADRL framework, adds a collaborative willingness network based on the original action value network of the agent, and uses the attention mechanism to improve the impact of the collaboration between agents on the action decision-making, while saving computing resources. The evaluation results show that, our CWN-MADRL algorithm can converge quickly, learn effective task recommendation strategies, and achieve better system performance compared with other benchmark algorithms.
AB - With the growth of maintenance market scale of automobile manufacturing enterprises, simple information technology is not enough to solve the problem of uneven resource allocation and low customer satisfaction in maintenance chain services. To solve this problem, this paper abstracts the automotive maintenance collaborative service into a multi-agent collaborative model based on the decentralized partially observable Markov decision progress (Dec-POMDP). Based on this model, a multi-agent deep reinforcement learning algorithm based on collaborative willingness network (CWN-MADRL) is presented. The algorithm uses a value decomposition based MADRL framework, adds a collaborative willingness network based on the original action value network of the agent, and uses the attention mechanism to improve the impact of the collaboration between agents on the action decision-making, while saving computing resources. The evaluation results show that, our CWN-MADRL algorithm can converge quickly, learn effective task recommendation strategies, and achieve better system performance compared with other benchmark algorithms.
KW - Deep reinforcement learning
KW - Equipment manufacturing
KW - Maintenance collaborative service
KW - Multi-agent
KW - Value decomposition
UR - http://www.scopus.com/inward/record.url?scp=85140452882&partnerID=8YFLogxK
U2 - 10.1007/978-3-031-16815-4_6
DO - 10.1007/978-3-031-16815-4_6
M3 - Conference contribution
AN - SCOPUS:85140452882
SN - 9783031168147
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 84
EP - 103
BT - Applied Cryptography and Network Security Workshops - ACNS 2022 Satellite Workshops, AIBlock, AIHWS, AIoTS, CIMSS, Cloud S and P, SCI, SecMT, SiMLA, Proceedings
A2 - Zhou, Jianying
A2 - Chattopadhyay, Sudipta
A2 - Adepu, Sridhar
A2 - Alcaraz, Cristina
A2 - Batina, Lejla
A2 - Casalicchio, Emiliano
A2 - Jin, Chenglu
A2 - Lin, Jingqiang
A2 - Losiouk, Eleonora
A2 - Majumdar, Suryadipta
A2 - Meng, Weizhi
A2 - Picek, Stjepan
A2 - Zhauniarovich, Yury
A2 - Shao, Jun
A2 - Su, Chunhua
A2 - Wang, Cong
A2 - Zonouz, Saman
PB - Springer Science and Business Media Deutschland GmbH
T2 - Satellite Workshops on AIBlock, AIHWS, AIoTS, CIMSS, Cloud S and P, SCI, SecMT, SiMLA 2022, held in conjunction with the 20th International Conference on Applied Cryptography and Network Security, ACNS 2022
Y2 - 20 June 2022 through 23 June 2022
ER -