Pedro Gomez-Gasquet¹, Alejandro Torres², Ana Esteso¹ and Maria Angeles Rodriguez¹

¹ Centro de Investigación de Gestión e Ingeniería de Producción (CIGIP). Universitat Politèc-nica de Valencia. Cno/ de Vera S/N, 46022 Valencia (Spain)

² Departamento de Organización de Empresas. Universitat Politècnica de Va-lencia. Cno/ de Vera S/N, 46022 Valencia (Spain).

pgomez@cigip.upv.es

Keywords: AI, machine learning, jobshop, flowshop, scheduling

1. Introduction

For almost 80 years the discipline of Operations Research (OR) has tackled production scheduling problems, incorporating in each decade the contributions that have been made from other areas (modeling, statistics, biology, etc.) with which the discipline has been growing and consolidating her body. After a strong start, Artificial Intelligence (AI) declined and it seemed not to be able to solve the returns that were being tested. However, in recent years it has been able to solve these aspects and at this time we could say that it seems to provide a different vision that is being incorporated in many disciplines. How could it be otherwise, the OR is choosing to incorporate approaches and techniques that come from AI in many of its papers.

This work is carried out review the important volume of scientific progress published in 2020 and early 2021 as a continuation of a previous one [1] in which the contributions that from an OR perspective included techniques clearly coming from AI were analyzed, where we already emphasized that most of the initial contributions were focused on Machine Learning, although the trend in the last years it was to use Reinforced Learning and Neuronal Networks. The set of scheduling problems that has been considered is wide since it includes any variation that addresses a configuration in jobshop or flowshop [2].

2. Conclusions and future research

This paper has made a review through the best contributions that have been found on a set of more than 100 references, all of them published in 2020. The publications selected in the period under review have not significantly changed the focus of interest in techniques that have been widely used for many years in the field of sequencing, such as neural networks or reinforced learning, in particular Q-learning algorithms. Although it may be somewhat circumstantial, we have seen that the number of publications in such a short period is high compared to those found in [1] in a much longer period. In any case, it can be concluded that the contribution of AI to OR is gradually consolidating the focus on Reinforced Learning and Neural Networks techniques, but that above all the use of Deep Q-Network is beginning to appear strongly. Due to the upward trend and the quality of the results obtained, everything seems to indicate that in the near future this is going to be a work area of maximum interest. The development of the general concept of deep learning has led to the adaptation and creation of methods and algorithms that experts are applying to the field of sequencing, especially in more complex problems such as jobshop scheduling.

After conducting this review and also considering the reality previously [1] the authors of this paper believe that the current IA techniques mentioned are relatively mature to be successfully applied to flowshop and jobshop problems. However, we believe that there is still a lot of work to be done in terms of modelling sequencing problems in the terms that IA needs. In other words, we believe that the IA methods analyzed as applied are not achieving the desirable results in terms of quality achieved in efficiency measures, nor in terms of computational speed, let alone simplicity. However, a new way of formulating the same sequencing problem in other terms would perhaps make it possible to move forward and correct the deficiencies we have discussed using the same IA techniques.

Acknowledgements. This research is being funded by project entitled NIOTOME (Ref. RTI2018-102020-B-I00) (MCI/AEI/FEDER, UE).

References

1. Gomez-Gasquet, P., Boza-Garcia, A., Navarro, A. (2020) Artificial intelligence for solving flowshop and jobshop scheduling problems: A literature review, 14th International Conference on Industrial Engineering and Industrial Management, XXIV Congreso de Ingeniería de Organización, Madrid, Spain, July 9-10, 2020 (publication in progress)
2. Brucker, P. (2007) Scheduling Algorithms, ed. 5, Springer-Verlag Berlin Heidelberg, doi:10.1007/978-3-540-69516-5
3. Medina-López, C., Marín-García, J.A., Alfalla-Luque, R. (2010) Una propuesta metodológica para la realización de Búsquedas sistemáticas de bibliografía. (A methodological proposal for the systematic literature review). Working Papers on Operations Management Vol 1, No 2 (13-30)