Integrating Machine Learning Into Vehicle Routing Problem: Methods and Applications

IEEE Access, Год журнала: 2024, Номер 12, С. 93087 - 93115

Опубликована: Янв. 1, 2024

The vehicle routing problem (VRP) and its variants have been intensively studied by the operational research community. existing surveys majority of published articles tackle traditional solutions, including exact methods, heuristics, meta-heuristics. Recently, machine learning (ML)-based methods applied to a variety combinatorial optimization problems, specifically VRPs. strong trend using ML in VRPs gap literature motivated us review state-of-the-art. To provide clear understanding ML-VRP landscape, we categorize related studies based on their applications/constraints technical details. We mainly focus reinforcement (RL)-based approaches because importance literature, while also address non RL-based methods. cover both theoretical practical aspects clearly addressing trends, gap, limitations advantages ML-based discuss some potential future directions.

Язык: Английский

---

Regression Machine Learning Models for the Short-Time Prediction of Genetic Algorithm Results in a Vehicle Routing Problem

World Electric Vehicle Journal, Год журнала: 2024, Номер 15(7), С. 308 - 308

Опубликована: Июль 14, 2024

Machine learning techniques have advanced rapidly, leading to better prediction accuracy within a short computational time. Such advancement encourages various novel applications, including in the field of operations research. This study introduces way utilize regression machine models predict objectives vehicle routing problems that are solved using genetic algorithm. Previous studies generally discussed how (1) research methods used independently generate optimized solutions and (2) values from given dataset. Some collaborations between fields as follows: input data for problems, optimize hyper-parameters models, (3) improve quality algorithms. differs types collaborative listed above. focuses on objective problem directly output data, without optimizing straightforward framework captures characteristics problem. The proposed is applied by generating algorithm then obtained values. numerical experiments show best random forest regression, generalized linear model with Poisson distribution, ridge cross-validation.

Язык: Английский

---

A contextual framework for learning routing experiences in last-mile delivery

Transportation Research Part B Methodological, Год журнала: 2025, Номер 194, С. 103172 - 103172

Опубликована: Фев. 18, 2025

Язык: Английский

---

Machine-learning component for multi-start metaheuristics to solve the capacitated vehicle routing problem

Applied Soft Computing, Год журнала: 2025, Номер 173, С. 112916 - 112916

Опубликована: Фев. 28, 2025

Язык: Английский

---

A non-anticipative learning-optimization framework for solving multi-stage stochastic programs

Annals of Operations Research, Год журнала: 2024, Номер unknown

Опубликована: Июль 3, 2024

Abstract We present a non-anticipative learning- and scenario-based prediction-optimization (ScenPredOpt) framework that combines deep learning, heuristics, mathematical solvers for solving combinatorial problems under uncertainty. Specifically, we transform neural machine translation frameworks to predict the optimal solutions of multi-stage stochastic programs. The learning models are trained efficiently using input solution data single-scenario deterministic problems. Then our ScenPredOpt creates mapping from inputs used in training into an output predictions close solutions. Non-anticipative Encoder-Decoder with Attention (NEDA) approach, which ensures non-anticipativity property programs and, thus, time consistency by calibrating learned information based on problem’s scenario tree adjusting hidden states network. In framework, percent predicted variables iteratively reduced through relaxation problem eliminate infeasibility. Then, linear relaxation-based heuristic is performed further reduce time. Finally, solver generate complete solution. results two NP-Hard sequential optimization uncertainty: multi-item capacitated lot-sizing multistage multidimensional knapsack. show can be factor 599 optimality gap only 0.08%. compare cutting-edge exact algorithms studied find more effective. Additionally, computational demonstrate solve instances larger number items scenarios than ones. Our learning-optimization approach beneficial programming involving binary solved repeatedly various types dimensions similar decisions at each period.

Язык: Английский

---

Reinforcement Learning Approach to Stochastic Vehicle Routing Problem With Correlated Demands

IEEE Access, Год журнала: 2023, Номер 11, С. 87958 - 87969

Опубликована: Янв. 1, 2023

We present a novel end-to-end framework for solving the Vehicle Routing Problem with stochastic demands (VRPSD) using Reinforcement Learning (RL). Our formulation incorporates correlation between through other observable variables, thereby offering an experimental demonstration of theoretical premise that non-i.i.d. provide opportunities improved routing solutions. approach bridges gap in application RL to VRPSD and consists parameterized policy optimized gradient algorithm generate sequence actions form solution. model outperforms previous state-of-the-art metaheuristics demonstrates robustness changes environment, such as supply type, vehicle capacity, correlation, noise levels demand. Moreover, can be easily retrained different scenarios by observing reward signals following feasibility constraints, making it highly flexible scalable. These findings highlight potential enhance transportation efficiency mitigate its environmental impact problems. implementation is available online.

Язык: Английский

---

Coefficient Prediction for Physically-based Cloth Simulation Using Deep learning

International Journal on Advanced Science Engineering and Information Technology, Год журнала: 2023, Номер 13(4), С. 1510 - 1517

Опубликована: Авг. 28, 2023

Physically-based cloth simulation involves modeling as a collection of particles or nodes connected by various types constraints. These interact with each other and the environment, such gravity collisions, to accurately simulate cloth's behavior. One essential component simulations is set material parameters coefficients that dictate physical properties, stiffness damping. Deep learning-based coefficient prediction in physically-based using machine learning techniques, specifically deep neural networks, predict from its geometric properties. The model trained dataset simulated instances, where are known. input properties cloth, dimensions, orientation, velocity. output best represent behavior under these conditions. This paper proposes method for predicting multi-label video classification approach. training data generated physics-based simulator, evaluated on some simulations, fabric falling down, collision, affected airflow. movement mass-spring-based simulation. results show transformer has much higher accuracy than models. study provides promising approach virtual simulations.

Язык: Английский

---

Learning generalizable heuristics for solving vehicle routing problem under distribution shift

Опубликована: Янв. 1, 2024

for their companionship and encouragement during my hard times, thank family

Язык: Английский

---

Combining Neighborhood Search with Path Relinking: A Statistical Evaluation of Path Relinking Mechanisms

Lecture notes in computer science, Год журнала: 2024, Номер unknown, С. 112 - 125

Опубликована: Янв. 1, 2024

Язык: Английский

---

A Contextual Framework for Learning Routing Experiences in Last-Mile Delivery

Опубликована: Янв. 1, 2024

Язык: Английский

---