Aprendizado de máquina para alocação de equipagem ferroviária

Loading...
Thumbnail Image

Journal Title

Journal ISSN

Volume Title

Publisher

Universidade Federal de Juiz de Fora (UFJF)

DOI

Abstract

Railway companies seek to optimize resources and reduce costs, which leads to becoming increasingly competitive in the market. A large portion of railway expenses is in the workforce’s payment, which, in addition to the salary, receives for hours of readiness, transport, food, and hotel expenses. Although there are specialized professionals whose function is to allocate the drivers more economically, many uncertainties exist at the time of the decision, leading many train drivers to carry out idle cycles. Machine learning techniques are ideal for generalizing new scenarios from a training database and can be applied to reduce uncertainties in many problems. The scientific literature shows that although there are many studies about the optimization of human resources in transport systems, this is still under development when it is necessary to use computational intelligence techniques, mainly in the railway environment. In this work, applied at MRS Logística S.A., a Brazilian railway operator, in a first step, exploratory data analysis techniques are used together with the rules generated by a decision tree to create and apply guidelines in crew allocation in the corporation. In the second stage, five machine learning algorithms learn the real need of a train driver for the rest of his cycle to reduce doubts at the time of his allocation, and may even lead to his dismissal from the service. Of the trained algorithms, in addition to the more classic ones such as neural network, support vector machine, decision tree, and random forest, ALMMo was used, a neuro-fuzzy, evolutionary and non-parametric model, which updates itself with the addition of new data. The results of the first stage showed a reduction of more than 50% in the occurrence of idle cycles in the company. In the second stage of the study, the algorithms showed an accuracy above 86%, on average, which meets all service levels established by the company. Finally, the decision tree, the random forest, and the ALMMo were considered suitable solutions for application in the company due to their performances and characteristics.

Description

Citation

Endorsement

Review

Supplemented By

Referenced By

Creative Commons license

Except where otherwise noted, this item's license is described as Acesso Aberto