Otimização da alocação de vagões no transporte logístico ferroviário integrando programação inteira mista e algoritmo genético

Abstract

This work presents a hybrid optimization model for the weekly allocation of flat wagons across the railway network of MRS Logística S.A. The primary objective is to minimize volume shortfalls (unmet demand), while a secondary objective focuses on reducing excessive fleet usage or idleness by applying a parameter, α, to the number of wagons. Initially, a Mixed-Integer Programming (MIP) model was employed—formulated in Pyomo and solved using the open-source solver CBC—which ensures optimal solutions for problems of moderate scale. To enhance scalability and experimental flexibility, a Genetic Algorithm (GA) was developed using the DEAP library; it was calibrated via grid search and incorporates repair routines to ensure the feasibility of the solutions. Input data—including weekly demand, availability by wagon subtype, effective capacities, and cycle times—were extracted from operational spreadsheets and processed using Python. In a historical case study covering June 2024 (five weeks), the MIP model met 465,383 tonnes of a total demand of 541,410 tonnes (85.96%), while the best-performing GA achieved 453,050 tonnes (83.68%) using 5,985 wagons. Processing times for both methods remained under a few minutes. These results demonstrate that open-source tools are capable of addressing complex logistical problems with near-optimal quality and competitive computational costs, offering a reproducible and economically advantageous methodology for railway operational planning.