Pasar al contenido principal

BUSQUEDA METAHEURISTICA CON APRENDIZAJE EN PROBLEMAS DE SCHEDULING SOSTENIBLE

PID2021-125919NB-I00

•
Nombre agencia financiadora Agencia Estatal de Investigación
Acrónimo agencia financiadora AEI
Programa Programa Estatal para Impulsar la Investigación Científico-Técnica y su Transferencia
Subprograma Subprograma Estatal de Generación de Conocimiento
Convocatoria Proyectos de I+D+i (Generación de Conocimiento y Retos Investigación)
Año convocatoria 2021
Plan Estatal 2021-2023 Plan Estatal de Investigación Científica y Técnica y de Innovación
Centro beneficiario UNIVERSITAT POLITÈCNICA DE VALÈNCIA

Resultados relacionados

Resultados totales (Incluyendo duplicados): 3
Encontrada(s) 1 página(s)

RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia
Publicación Artículo científico (JournalArticle)

Driving energy efficiency in scheduling manufacturing: An advanced instance generation and a metaheuristic approach

  • Pérez-Bernal, Christian|||0000-0002-9121-7939
  • March, C
  • Miguel A. Salido|||0000-0002-4835-4057
[EN] In pursuit of sustainability, modern manufacturing faces complex scheduling challenges to optimize energy, labor, and machine utilization. Many such problems can be formulated as a Job-Shop Scheduling Problem (JSP), augmented with sustainability requirements. We propose an instance generation framework for JSP incorporating energy-related factors (notably machine speed scaling and associated energy consumption) to enable realistic and diverse benchmarking. The generator uses probabilistic distributions to create complex scenarios mirroring real-world conditions, allowing robust evaluation of scheduling algorithms under varying machine speeds, processing times, and energy demands. In addition, we develop a novel metaheuristic for energy-efficient scheduling: a genetic algorithm (GA) enhanced with machine-learning-based adaptation. The GA employs a task-based representation that treats machine speed as a decision variable, optimizing the trade-off between makespan, energy usage, and tardiness. Experimental results demonstrate that the proposed GA significantly outperforms state-of-the-art exact solvers (e.g., GUROBI, CPLEX, GECODE) on larger problem instances. While exact solvers find optimal solutions for small instances, they struggle to scale; in contrast, our GA consistently achieves near-optimal solutions efficiently for large-scale, energy-aware scheduling problems., The authors gratefully acknowledge the financial support of the European Social Fund (Investing In Your Future), the Spanish Ministry of Science (project PID2021-125919NB-I00).
RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia
Publicación Artículo científico (JournalArticle)

A hybrid metaheuristic with learning for a real supply chain scheduling problem

  • Pérez-Bernal, Christian|||0000-0002-9121-7939
  • Climent Aunes, Laura Isabel
  • Miguel A. Salido|||0000-0002-4835-4057
  • Nicoló, Giancarlo
  • Arbelaez, Alejandro
[EN] In recent decades, research on supply chain management (SCM) has enabled companies to improve their environmental, social, and economic performance.This paper presents an industrial application of logistics that can be classified as an inventory-route problem. The problem consists of assigning orders to the available warehouses. The orders are composed of items that must be loaded within a week. The warehouses provide an inventory of the number of items available for each day of the week, so the objective is to minimize the total transportation costs and the costs of producing extra stock to satisfy the weekly demand. To solve this problem a formal mathematical model is proposed. Then a hybrid approach that involves two metaheuristics: a greedy randomized adaptive search procedure (GRASP) and a genetic algorithm (GA) is proposed. Additionally, a meta-learning tuning method is incorporated into our hybridized approach, which yields better results but with a longer computation time. Thus, the trade-off of using it is analyzed.An extensive evaluation was carried out over realistic instances provided by an industrial partner. The proposed technique was evaluated and compared with several complete and incomplete solvers from the state of the art (CP Optimizer, Yuck, OR-Tools, etc.). The results showed that our hybrid metaheuristic outperformed the behavior of these well-known solvers, mainly in large-scale instances (2000 orders per week). This hybrid algorithm provides the company with a powerful tool to solve its supply chain management problem, delivering significant economic benefits every week., The authors gratefully acknowledge the financial support of the European Social Fund (Investing In Your Future) , the Spanish Ministry of Science (project PID2021-125919NB-I00), and valgrAI-Valencian Graduate School and Research Network of Artificial Intelligence and the Generalitat Valenciana, Spain, and co-funded by the European Union. The authors also thank the industrial partner Logifruit for its support in the problem specification and the permission to generate randomized data for evaluating the proposed algorithms.
RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia
Publicación Artículo científico (JournalArticle)

Optimizing Energy Efficiency in Unrelated Parallel Machine Scheduling Problem through Reinforcement Learning

  • Pérez-Bernal, Christian|||0000-0002-9121-7939
  • Miguel A. Salido|||0000-0002-4835-4057
  • March, C
[EN] The industrial sector plays a significant role in global energy consumption and greenhouse gas emissions. To reduce this environmental impact, it's crucial to implement energy-efficient manufacturing systems that utilize sustainable materials and optimize energy usage. This can lead to benefits such as reduced carbon footprints and cost savings. In recent years, metaheuristic approaches have been focused on minimizing energy consumption within the Unrelated Parallel Machine Scheduling Problem (UPMSP). Traditional methods often overlook complex factors like release dates, due dates, and job setup times. This research introduces a novel algorithm that integrates reinforcement learning (RL) with a genetic algorithm (GA) to address this gap. The proposed RLGA algorithm, rooted in the dynamic field of evolutionary reinforcement learning, breaks down policies into smaller components to isolate essential parameters for problem-solving. Through comprehensive analysis, hyperparameters that influence optimal results are identified, facilitating automated hyperparameter selection and optimization. The expert system takes into account problem characteristics such as machine or job saturation, job overlap, and the maximum values of target variables, allowing instances to be grouped into clusters. These clusters are solved using a genetic algorithm with varying combinations of mutation and crossover hyperparameters. The most suitable approach for each cluster is determined by analyzing the results, and this configuration of hyperparameters is applied iteratively to optimize the solution search. The effectiveness of RLGA is evaluated across benchmark instances with different complexities, machine sets, jobs, and constraints. Comprehensive comparisons against existing methods highlight the superior performance and efficiency of RLGA in optimizing energy use and solution quality. Experimental results show that RLGA outperforms well-known solvers like CPO, CPLEX, OR-tools, and Gecode, making it a promising approach for optimizing energy-efficient manufacturing systems., This work was supported by grant PID2021-125919NB-I00 funded by MICIU/AEI/10.13039/501100011033/FEDER, UE