Master's Thesis

A Performance Analysis of Reinforcement Learning Applications in Multi-Echelon Inventory Optimization

Francisco Maria Correia Isidro — 2025

Key information

Authors:

Francisco Maria Correia Isidro (Francisco Maria Correia Isidro)

Supervisors:

Cátia Rafaela Ferreira Medeiros da Silva Coelho (Cátia Rafaela Ferreira Medeiros da Silva); Miguel Vieira

Published in

December 2, 2025

Abstract

Effective inventory management is a cornerstone of modern supply chain operations, increasingly challenged by market volatility and demand uncertainty. In this context, multi-echelon inventory optimization remains a critical challenge where traditional approaches face significant limitations. While Deep Reinforcement Learning (DRL) is a promising adaptive methodology, limited research has examined how its performance compares against a theoretical optimum across varying scales of system complexity. This thesis aims to fill this gap by empirically quantifying this performance gap to assess the conditions under which DRL generates near-optimal inventory policies. To achieve this, the performance of policies derived from a suite of DRL algorithms is evaluated against the theoretical optimum established by a deterministic Linear Programming (LP) model with perfect foresight. The analysis is conducted across four scenarios of escalating scale and structural complexity. The results reveal that as system complexity increases, DRL policies converge to near-optimality. This is attributed to their ability to implicitly learn and exploit system-wide dynamics. The study also demonstrates the superior performance and sample efficiency of off-policy algorithms over on-policy counterparts. Leading DRL agents learned robust and computationally instantaneous policies. Overall, the research underscores the potential of DRL to complement traditional optimization techniques, offering valuable insights for managing complex and uncertain supply chains.

Publication details

Authors in the community:

Supervisors of this institution:

Fields of Science and Technology (FOS)

other-engineering-and-technologies - Other engineering and technologies

Publication language (ISO code)

eng - English

Rights type:

Embargo lifted

Date available:

September 17, 2026

Institution name

Instituto Superior Técnico