Bruno Pimentel
Research Subjects
Bruno Pimentel has a total of 8 co-authors. Here are the top co-authors by number of shared publications:
Eguinaldo de Souza
1 shared publicationFernanda Alves
1 shared publicationGeraldo Robson Mateus
1 shared publicationGlauber Bomtempo
1 shared publication- Júlio César Alves
Universidade Federal de Lavras
1 shared publication Martin Ravetti
1 shared publication
Latest Publications
- Open Access
- 20 April 2026
- Published by Elsevier in SSRN Electronic Journal
A Hybrid Mixed-Integer Linear Programming and Reinforcement Learning Framework for Large-Scale Optimization
This paper proposes a hybrid optimization framework that integrates Mixed-Integer Linear Programming (MILP) with Reinforcement Learning (RL) to address large-scale, dynamic production planning problems. The framework is conceived as an interconnected architecture composed of three modules: a MILP modeling module that ensures global consistency and feasibility, a learning module based on Proximal Policy Optimization (PPO) that learns adaptive decision policies, and an evaluation and feedback module that iteratively refines both components. Additionally, the framework targets problems with quasi-decomposable structures, where local decisions are tightly coupled through global balance constraints. The proposed architecture is evaluated using a real-world case study from Samarco, a complex iron ore production system involving long-distance pipeline transport and interdependent operational stages. Computational experiments using real and simulated data demonstrate that the hybrid framework achieves faster convergence and improved adaptability compared to standalone MILP and RL approaches, offering a scalable and intelligent solution for industrial production planning.
- 20 April 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessA Hybrid Mixed-Integer Linear Programming and Reinforcement Learning Framework for Large-Scale Optimization
This paper proposes a hybrid optimization framework that integrates Mixed-Integer Linear Programming (MILP) with Reinforcement Learning (RL) to address large-scale, dynamic production planning problems. The framework is conceived as an interconnected architecture composed of three modules: a MILP modeling module that ensures global consistency and feasibility, a learning module based on Proximal Policy Optimization (PPO) that learns adaptive decision policies, and an evaluation and feedback module that iteratively refines both components. Additionally, the framework targets problems with quasi-decomposable structures, where local decisions are tightly coupled through global balance constraints. The proposed architecture is evaluated using a real-world case study from Samarco, a complex iron ore production system involving long-distance pipeline transport and interdependent operational stages. Computational experiments using real and simulated data demonstrate that the hybrid framework achieves faster convergence and improved adaptability compared to standalone MILP and RL approaches, offering a scalable and intelligent solution for industrial production planning.