M.A. Kazantsev1, I.A. Pinchuk2, D.G. Gaifulin3, E.E. Noskova4

1-4 Joint-Stock Company «Special-Purpose Enterprise «Radiosvyaz» (Krasnoyarsk, Russia)

1 mkaz@mail.ru, 2 pinchuk.ivan@yandex.ru, 3 me@dgayfulin.ru, 4 een90@mail.ru

Problem Statement. The job shop scheduling problem (JSSP) is a combinatorial optimization problem of non -polynomial (NP-hard) complexity, in which it is necessary to assign the processing order of jobs on a limited number of machines so as to minimize the total schedule length (makespan). Traditional methods (exact algorithms and various heuristics) often prove insufficient in terms of solution quality and computational efficiency when solving large-scale problems, and they adapt poorly to dynamically changing production conditions.

Objective. To explore the applicability of modern deep learning methods, particularly reinforcement learning (RL), for automating the scheduling process. To compare the schedules generated by deep learning –based methods with those obtained using classical approaches, and to evaluate the potential of neural network technologies for solving scheduling problems in industrial environments.

Results. Several neural network architectures were developed and implemented, incl uding models based on multilayer perceptrons (Learning to Dispatch, L2D), graph neural networks (Graph Scheduler for Production, GraSP-RL), Petri net–based networks (PetriRL), and a hybrid model combining reinforcement and imitation learning elements (Hybr idIL). The Proximal Policy Optimization (PPO) algorithm was used for training, enhanced with an action -masking mechanism and restriction of action selection to the top -k candidates according to the Shortest Processing Time (SPT) rule. The trained agents de monstrated the ability to generate production schedules with reduced machine idle times and total makespan comparable to those obtained by heuristic algorithms. Moreover, the neural networks produced schedules significantly faster than conventional optimization methods, which is especially important for intra-shop planning in shift and daily scheduling tasks.

Practical Significance. The obtained results demonstrate the potential of RLbased approaches for automating production scheduling. Further research is required to improve model scalability, incorporate real-world production constraints, and enhance training quality. However, the developed methods can already serve as a foundation for integrating intelligent scheduling systems into industrial solutions, thereby improving the efficiency and flexibility of production management.

Kazantsev M.A., Pinchuk I.A., Gaifulin D.G., Noskova E.E. Application of neural network methods to job shop scheduling problem. Dynamics of complex systems. 2026. V. 20. № 1. P. 18−29. DOI: 10.18127/j19997493-202601-02 (in Russian).

1. Garey M. R., Johnson D. S., Sethi R. The Complexity of Flowshop and Jobshop Scheduling, Mathematics of Operations Research. 1976. V. 1. № 2. P. 117–129.
2. Kazantsev M.A., Pinchuk I.A., Noskova E.E. The planning-optimization level of digitalecosystem development in radio-electronic manufacturing. Dynamics of Complex Systems – XXI century. 2025. V. 19. № 1. P. 14–23. https://doi.org/10.18127/j19997493-202501-02 (in Russian).
3. Lv L., Zhang C., Fan J., Shen W. Deep Reinforcement Learning for Job Shop Scheduling Problems: A Comprehensive Literature Review. Knowledge-Based Systems. 2025. V. 321. № 4.
4. Waschneck B., Reichstaller A., Belzner L., Altenmüller T., Bauernhansl T., Knapp A., Kyek A. Optimization of Global Production Scheduling with Deep Reinforcement Learning. Procedia CIRP. 2018. V. 72. Р. 1264–1269.
5. Rinciog A., Meyer A. Towards Standardising Reinforcement Learning Approaches for Production Scheduling Problems. Procedia CIRP. 2022. V. 107. Р. 1112–1119.
6. Yamada T., Nakano R. Job-Shop Scheduling. IEE Control Engineering Series. 1997. № 55. Р. 134–160.
7. Zhang C., Song W., Cao Z., Zhang J., Tan P. S., Xu C. Learning to Dispatch for Job Shop Scheduling via Deep Reinforcement Learning. 34th Conference on Neural Information Processing Systems. 2020.
8. Abdul Hameed M., Schwung A. Graph Neural Networks-Based Scheduler for Production Planning Problems Using Reinforcement Lear ning. Journal of Manufacturing Systems. 2023. V. 69. Р. 91–102.
9. Sun H., Hao Zh., Hu Z., Lv J., Wang C. A Rescheduling Strategy Based on Deep Reinforcement Learning for New Order Insert ion. Engineering Applications of Artificial Intelligence. 2025. V. 159. 1
10. Khadivi M., Charter T., Yaghoubi M., Ahang M., Shojaeinasab A., Najjaran H. Deep Reinforcement Learning for Machine Scheduling: Methodology, the State-of-the-Art, and Future Directions. Computers & Industrial Engineering. 2025. V. 200. 1
11. Lassoued S., Schwung A. Introducing PetriRL: An Innovative Framework for JSSP Resolution Integrating Petri Nets and Event-Based Reinforcement Learning. Journal of Manufacturing Systems. 2024. V. 74. Р. 690–702. 1
12. Lassoued S., Baheti L., Weiß -Borkowski N., Lier S., Schwung A. Collaborative Dynamic Scheduling in a SelfOrganizing Manufacturing System Using Multi-Agent Reinforcement Learning. Journal of Manufacturing Systems. 2025. V. 82. Р. 405– 4 1
13. Application of neural network methods to job shopscheduling problem Dynamics of complex systems / Dinamika slozhnykh sistem, V. 20, № 1, 2026, p. 18−29 29 1
14. Schulman J., Wolski F., Dhariwal P., Radford A., Klimov O. Proximal Policy Optimization Algorithms [Electronic resource]. OpenAI: [website]. [2017]. Access mode: https://openai.com/index/openai-baselines-ppo/ (accessed: 6 October 2025). 1
15. Kazantsev M.A., Pinchuk I.A., Noskova E.E. Program for Generating Initial Data for Optimization Models of Production Planning, certificate No 2025682067, Russian Federation, filed 08 Jul 2025, published 20 Aug 2025, applicant JSC “Research and Production Enterprise ‘Radiosvyaz’” 1
16. Galeev R.G., Kapulin D.V., Kazantsev M.A. Production Logistics of an Instrument-Making Enterprise: Textbook. Krasnoyarsk: Siberian Federal University. 2021. 265 p. (in Russian).