Consensus management and multi-agent reinforcement learning in the problems of structuring project networks

An approach to the construction of technological platforms (project networks) designed to enable self-organization of participants with key competencies into a team to carry out activities with initially set goals, the achievement of which determines the completion of the project, is considered. At the initial stage, the project network is in ‘the sleep mode’, i.e. the usual for a social network information exchange between potential project team members takes place on the network and, consequently, ‘traditional agents’ or actors interact on the network. A two-level scheme is proposed for organizing the process of interaction between agents of the project network in work teams: intra-cluster and inter-cluster. The effectiveness of the first interaction is estimated as the result of consensus modeling in asynchronous multi-agent systems with discrete and continuous time. At the same time, if consensus is reached, then the cluster at the second level of the hierarchy can be considered as a single agent node participating in the next cycle of interaction – inter-cluster. At this level, the solutions being formed are considered as Markov decision-making processes. Accordingly, as a mathematical apparatus for modeling this type of interaction, it is planned to use one of the machine learning methods – reinforcement learning when solving the problem of optimal resource allocation between processes within a single project.

1. 1. Timofeev K.N. Project networks. In: Innovation management: from theory to practice. Proceedings of VII annual (II international) scientific and practical conference at Management Department. Saint-Petersburgh, OOP NIU VSHE – Saint-Petersburgh; 2012:127–135. (In Russ.).

2. 2. Kataev A.V., Kataeva T.M. Project managements based on the dynamic affiliate network: a monograph. Rostov-on-Don – Taganrog, Publishing House of Southern Federal University; 2017. 125 p. (In Russ.).

3. 3. Voronina L.A., Ratner S.V. Russian scientific and innovation networks: experience, issues, prospects. Moscow, INFRA-M; 2010. 254 p. (In Russ.).

4. 4. Gubanov D.A., Novikov D.A., Chkhartishvili A.G. Social networks: models of informational influence, management and confrontation. Moscow, PHIZMATLIT; 2010. 228 p. (In Russ.).

5. 5. Mengbin Ye, Ji Liu, Lili Wang, Brian D.O. Anderson, Ming Cao. Consensus and disagreement of heterogeneous belief systems in influence networks; 2018. Available from: https://arxiv.org/abs/1812.05138 (accessed on: 30.11.2022).

6. 6. Proskurnikov A.V. Usrednjajushhie algoritmy i neravenstva v zadachah mnogoagentnogo upravlenija i modelirovanija Saint-Petersburgh: Saint-Petersburgh State University; 2021. Available from: https://disser.spbu.ru/files/2021/disser_proskurnikov.pdf (accessed on 28.11.2022).

7. 7. Parsegov S.E. Algoritmy upravlenija formaciej v zadache ravnomernogo raspolozhenija agentov. Author’s abstract. Moscow, 2013; 22 p. (In Russ.).

8. 8. Sutton R.S., Barto A.J. Reinforcement Learning: An Introduction. 2nd ed. / translated from English by A.A. Slinkin. Moscow, DMK Press; 2020. 552 p. (In Russ.).

9. 9. Mnih V., Kavukcuoglu K., Silver D., Graves A., Antonoglou I., Wierstra D., Riedmiller M. Playing Atari with deep reinforcement learning. 2013. Available from: https://arxiv.org/abs/1312.5602v1 (accessed on 30.11.2022).

10. 10. Reinforcement Learning Toolbox Documentation. Available from: https://www.mathworks.com/help/reinforcement-learning/ (accessed on 28.11.2022).