Last-mile delivery problems where trucks and drones collaborate to deliver goods to final customers are considered. We focus on settings where a fleet with several homogeneous trucks work in parallel to collaborative drones, able to combine with each other to optimize speed and power consumption for deliveries. A heuristic for the min-max vehicle routing problem is coupled with constraint programming models, leading to an effective method able to provide several state-of-the-art solutions for the instances commonly adopted in the literature.
Julia is a programming language suitable for data analysis and scientific computing that combines simplicity of productivity languages with characteristics of performance-oriented languages. In this paper, we are interested in studying the use of Julia to implement Multi-Objective MetaHeuristics. Concretely, we use the Java-based jMetal framework as a reference support and investigate how Julia could be used to design and develop the component-based architecture for multi-objective evolutionary algorithms that jMetal provides. By using the NSGA-II algorithm as an example, we analyze the advantages and shortcomings of using Julia in this context, including aspects related to reusing jMetal code and a performance comparison.
This work proposes a Large Neighborhood Search Metaheuristic for solving a mixed-model assembly line balancing problem with walking workers and dynamic task assignment. The considered problem is a multi-stage stochastic program with integer recourse. These problems are very hard to solve because the number of binary variables increases exponentially with the number of production cycles. We study different decomposition approaches, and our results suggest that re-optimizing for a sub-tree outperforms other decompositions, such as model-based or station decomposition.
With the escalating demand for meal delivery services, this study delves into the Meal Delivery Routing Problem (MDRP) within the context of last-mile logistics. Focusing on the critical aspects of courier allocation and order fulfillment, we introduce a novel approach utilizing a GRASP metaheuristic. The algorithm optimizes the assignment of couriers to orders, considering dynamic factors such as courier availability, order demands, and geographical locations. Real-world instances from a Colombian delivery app form the basis of our computational analysis. Calibration of GRASP parameters reveals a delicate trade-off between solution quality and computational time. Comparative results with a simulation-optimization based study underscore GRASP’s competitive performance, demonstrating strengths in fulfilling orders and routing efficiency across diverse instances. This research enhances operational efficiency in the burgeoning food delivery industry, shedding light on practical algorithms for last-mile logistics optimization.
We study a general class of single-machine scheduling problems with setup time/cost and no idle time in the schedule. It includes a variety of other scheduling and routing problems as special cases. A polynomial size mixed-integer linear programming (MILP) formulation is presented. A tabu search (TS) algorithm, built upon a novel composite-move neighborhood structure, is developed for solving the addressed problem effectively and efficiently.
Metaheuristic hybrids equipped with multiple restarts have shown promise in complex optimization problems. A critical challenge in this domain, particularly for exploration-only exploitation-only hybrids, is determining optimal transition points between algorithms and restart locations. Each component of these hybrids excels in a specific task but may underperform in others, making transition and restart decisions crucial. This paper introduces an innovative solution to these challenges using reinforcement learning. We apply this approach to the UES-CMAES hybrid, training reinforcement learning agents to intelligently manage algorithm transitions and restarts. Evaluation on the CEC’13 benchmark suite demonstrates the efficacy of this method, indicating significant improvements in optimization performance. Our findings not only confirm the potential of reinforcement learning in enhancing metaheuristic hybrids but also pave the way for new research directions in intelligent optimization strategies.
In the competitive passenger air transport market, low-cost airlines continue strengthening their position, contrasting sharply with traditional carriers. This article delves into the unique operational strategies of these airlines, focusing on their reliance on ancillary services. Among these services, seat selection stands out as a crucial revenue enhancer. The study emphasizes the importance of low-cost carriers ensuring the availability of specific seats for direct purchase, thereby avoiding their allocation through automatic seat assignment algorithms, commonly activated for passengers who do not opt for specific seating. A notable consumer behavior observed is the preference for passengers on the same booking to be seated together. Low-cost airlines can capitalize on this trend by encouraging seat purchases and using automated seat assignments to strategically separate passengers traveling together unless they opt for paid seat selection. This work presents a novel approach to the seat assignment problem based on a GRASP algorithm; this approach is beneficial due to its low requirement for extensive parameter calibration, intuitive nature, and adaptability to different airline scenarios. Using an actual flight database of a low-cost Colombian airline, we have compared the airline’s rule-based heuristics, a network flow model, and our metaheuristic approach; the results obtained are satisfactory in terms of solution quality and computational cost. The proposed solution offers a viable, cost-effective alternative to specialized software solutions, aligning with the financial constraints typical of low-cost carriers while effectively enhancing their seat assignment process to optimize revenue generation.
Particle Swarm Optimization (PSO) is one of the most famous swarm-based algorithms used for solving optimization problems. PSO has received growing attention within many fields of the research community. Since its inception, some prominent improvements have been created. Within the broad spectrum of proposals that have emerged in the last few decades, improvements have been made to swarm initialization; new parameters have been introduced, such as the constraint on the inertia weight coefficient, and even mutation operators have been introduced to the PSO. However, the PSO has drawbacks and shortcomings, such as lack of convergence, loss of diversity, or stagnation at local minima. This paper proposes a population-based approach to hyperparametric solutions; the central premise is that each of the swarm particles has different parameters so that each has unique characteristics to promote exploitation-exploration and guide a heuristic with healthy diversity; empirical analysis and statistical tests performed on the proposed algorithm show the feasibility of the approach compared to improved versions of PSO found in the literature.
As digitalization spreads across the globe, the amount of information available is increasing exponentially and users are suffering from information overload. Recommender systems present a feasible and effective means to guide and expose users to products and items which align with their preferences. Specifically with the boom of social networks, collaborative filtering recommender systems offer a means to suggest highly relevant items to a user based on their shared interests with other users in the system. Despite major advancements through the integration of machine learning and hybrid systems, collaborative filtering algorithms struggle to handle large and sparse datasets which hampers the system’s ability to provide accurate recommendations. Metaheuristic techniques have been successful in improving collaborative filtering recommender systems despite data size and sparsity. This study presents a review of different attempts to optimize collaborative filtering recommender systems inclusive of metaheuristic techniques in this evolution which highlights an evident gap in standardized evaluation metrics of recommender systems.
This work leverages reinforcement learning for designing a new variant of Construct, Merge, Solve and Adapt (CMSA), a rather new hybrid metaheuristic for combinatorial optimization. We demonstrate a twofold improvement over the standard CMSA. Firstly, the new variant simplifies CMSA by eliminating the need for a greedy function to probabilistically generate solutions. Additionally, it performs better, as we demonstrate in the context of the Minimum Dominating Set (MDS) problem.
Event-interval sequences are defined as multivariate series of events that occur over time. The classification of event-interval sequences has gained increasing attention among researchers in the field of time series analysis due to their broad applicability, as for instance in healthcare and weather forecasting. This paper focuses on the optimized extraction of interpretable features from event-interval sequences to construct supervised classifiers. The current state-of-the-art is represented by e-lets, which are randomly sampled subsequences of event-intervals. We propose a new approach to interpretable classification of event-interval sequences based on sparse-lets, a novel generalization of e-lets. Our approach relies on genetic algorithms to learn sparse-lets, generating optimized and interpretable features. We evaluate the performance of our method through experiments conducted on benchmark datasets, and compare it against the state-of-the-art. Computational results show that our method is a viable competitor in terms of classification accuracy. Moreover, we show that our method generates simpler features than competing approaches, retaining only the most important information.
Search Trajectory Networks (STNs) are visualizations of directed graphs designed to analyze the behavior of stochastic optimization algorithms such as metaheuristics. Their purpose is to provide researchers with a tool that allows them to gain a deeper understanding of the behavior exhibited by multiple algorithms when applied to a specific instance of an optimization problem. In this short paper, we present two examples of our work in which STN graphics have helped us to discover interesting and useful algorithm/problem characteristics.
With the increasing availability of data and the development of powerful algorithms, deep neural networks have become an essential tool for all sectors. However, it can be challenging to automate the process of building and tuning them, due to the rapid growth of data and their complexity. The demand for handling large amounts of data has led to an increasing number of hidden layers and hyperparameters. A framework or methodology to design the architecture of deep neural networks will be crucial in the future, as it could significantly speed up the process of using deep learning models. We present here a first attempt to create an algorithm that combines aspects of Neural Architecture Search and Hyperparameter Optimization to build and optimize a neural network architecture. The particularity of our algorithm is that it is able to learn how to link neural layers of different types to create increasingly performant neural network architectures. We conducted experiments on four different tasks, including regression, binary and multi-classification, and forecasting, to compare our algorithm with common machine learning models.
In this study, we propose a scheduling problem that stems from technological disasters, characterized by risks and the dynamic evolution of hazardous products. The goal is to clean or neutralize these hazardous products to reduce the overall risk of contamination spreading throughout the environment, affecting inhabitants and agricultural areas. The problem is approached as a Resource-Constrained Project Scheduling with Risk and Product Transformation Dynamics. We present a mathematical formulation and introduce an Iterated Local Search metaheuristic. We fine-tune the metaheuristic parameters using a machine learning package and conduct several numerical experiments to assess performance and gain insights into this innovative application.
This paper introduces an extended version of the min-Knapsack problem with compactness constraints (mKPC). The idea is to define penalty values for certain items when they are not selected in the knapsack. In addition to cost, weight, and compactness constraints in the mKPC, which require selected items to remain within close proximity, the min-Knapsack problem with compactness constraints and penalty values (mKPCP) incorporates penalty values for excluding certain items. The method outlined in this study leverages the learning mechanism of a metaheuristic approach, Fixed Set Search, integrating it with integer programming to address partial solutions throughout the process. To enhance the learning mechanism process, the initial population of solutions is generated through an algorithm that randomly creates solutions considering the compactness constraint and the item sequences, with the aim of enhancing diversity. New instances are proposed to evaluate the proposed method on the mKPCP. The method is also tested on the mKCP to compare with existing methods. The experiments indicate that the proposed method yields promising outcomes across a diverse set of instances. The approach does not rely heavily on any unique characteristics of the problem and could be adapted to other binary problems, including the minimum vertex cover problem and the facility location problem, with small adjustments.
This study explores the dynamic configuration of a population-based metaheuristic with reinforcement learning. Beyond achieving high performance, our dual focus involves utilizing hyperparameters as indicators for transitions between exploration and exploitation phases. We investigate how this information can be effectively harnessed for responsive balance tailored to each problem instance. Specifically, we analyze the potential of integrating the Local Optima Network (LON), an abstraction of the fitness landscape, to inform parameter generation. To study the relationship between indicators and responsive control, we embed the algorithm within a reinforcement learning framework.
This study introduces an innovative optimization technique rooted in hybridizing the Golden Sine Algorithm II and the Non - Monopolized Search algorithm tailored to address unconstrained problems. The core concept underlying Golden Sine Algorithm II hinges on leveraging the diminishing pattern of the sine function and the golden ratio to navigate the solution landscape effectively; meanwhile, the Non-Monopolized Search is employed to improve the exploitation as a local search mechanism. Our proposal is called improved Golden Sine Algorithm II with Non-Monopolized Local Search (GSII-LS). Notably, GSII-LS is designed to complement and enhance existing optimization methodologies, working in synergy with non-monopolizing search strategies. To assess its efficacy, GSII-LS is subjected to rigorous testing across 34 benchmark functions for unconstrained optimization. Comparative analysis against optimization algorithms is conducted using established evaluation criteria. Results demonstrate that GSII-LS consistently achieves superior convergence towards global optima across numerous benchmark functions.
This paper extends with matheuristic operators the seminal DSATUR heuristic for the Vertex Coloring Problem. Firstly, matheuristics are proposed to initialize saturation computing using a clique, a partial optimal coloring with selected vertices or combining both previous strategies. Secondly, an Integer Linear Programming formulation is designed to have larger local greedy optimization in DSATUR construction scheme. Thirdly, dual bounds are obtained with local optimization to improve first lower bounds implied by cliques. Computational results are provided to analyze inefficiency causes of DSATUR heuristic, highlighting the strengths and weaknesses of DSATUR heuristics.
This paper investigates a specific instance of the static repositioning problem within station-based bike-sharing systems. Our study incorporates operational and damaged bikes, a heterogeneous fleet, and multiple visits between stations and the depot. The objective is to minimize the weighted sum of the deviation from the target number of bikes for each station, the number of damaged bikes not removed, and the total time used by vehicles. To solve this problem, we propose a matheuristic approach based on a randomized multi-start algorithm integrated with an integer programming model for optimizing the number of operatives and damaged bikes that will be moved between stations and/or the depot (loading instructions). The algorithm's effectiveness was assessed using instances derived from real-world data, yielding encouraging results. Furthermore, we adapted our algorithm to a simpler problem studied in the literature, achieving competitive outcomes compared to other existing methods. The experimental results in both scenarios demonstrate that this algorithm can generate high-quality solutions within a short computational time.
In this paper the fixed set search (FSS), a population-based metaheuristic, is applied to the Maximum Disjoint Dominating Sets Problem (MDDSP). Initially, a greedy randomized adaptive search procedure (GRASP) is developed to solve the MDDSP. Subsequently, the FSS enhances GRASP by incorporating a learning mechanism that identifies common elements in high-quality solutions. Computational experiments show that the proposed method significantly outperforms current state-of-the-art methods.
