Laura Calvet

Waste collection under uncertainty: a simheuristic based on variable neighbourhood search

Ongoing population growth in cities and increasing waste production has made the optimisation of urban waste management a critical task for local governments. Route planning in waste collection can be formulated as an extended version of the well-known vehicle routing problem, for which a wide range of solution methods already exist. Despite the fact that real-life applications are characterised by high uncertainty levels, most works on waste collection assume deterministic inputs. In order to partially close this literature gap, this paper first proposes a competitive metaheuristic algorithm based on a variable neighbourhood search framework for the deterministic waste collection problem. Then, this metaheuristic is extended to a simheuristic algorithm in order to deal with the stochastic problem version. This extension is achieved by integrating simulation into the metaheuristic framework, which also allows a closer risk analysis of the best-found stochastic solutions. Different computational experiments illustrate the potential of our methodology. [Received: 13 January 2016; Revised: 25 April 2016; Revised: 19 September 2016; Revised: 18 October 2016; Accepted: 25 October 2016]

A statistical learning based approach for parameter fine-tuning of metaheuristics

Metaheuristics are approximation methods used to solve combinatorial optimization problems. Their performance usually depends on a set of parameters that need to be adjusted. The selectionof appropriate parameter values causes a loss of efficiency, as it requires time, and advanced analytical and problem-specific skills. This paper provides an overview of the principal approaches to tackle the Parameter Setting Problem, focusing on the statistical procedures employed so far by the scientific community. In addition, a novel methodology is proposed, which is tested using an already existing algorithm for solving the Multi-Depot Vehicle Routing Problem.

Designing e-commerce supply chains: a stochastic facility–location approach

e-Commerce activity has been increasing during recent years, and this trend is expected to continue in the near future. e-Commerce practices are subject to uncertainty conditions and high variability in customers’ demands. Considering these characteristics, we propose two facility–location models that represent alternative distribution policies in e-commerce (one based on outsourcing and another based on in-house distribution). These models take into account stochastic demands as well as more than one regular supplier per customer. Two methodologies are then introduced to solve these stochastic versions of the well-known capacitated facility–location problem. The first is a two-stage stochastic-programming approach that uses an exact solver. However, we show that this approach is not appropriate for tackle large-scale instances due to the computational effort required. Accordingly, we also introduce a “simheuristic” approach that is able to deal with large-scale instances in short computing times. An extensive set of benchmark instances contribute to illustrate the efficiency of our approach, as well as its potential utility in modern e-commerce practices.

Minimizing Trigger Error in Parametric Earthquake Catastrophe Bonds via Statistical Approaches

The insurance and reinsurance industry, some governments, and private entities employ catastrophe (CAT) bonds to obtain coverage for large losses induced by earthquakes. These financial instruments are designed to transfer catastrophic risks to the capital markets. When an event occurs, a Post-Event Loss Calculation (PELC) process is initiated to determine the losses to the bond and the subsequent recoveries for the bond sponsor. Given certain event parameters such as magnitude of the earthquake and the location of its epicenter, the CAT bond may pay a fixed amount or not pay at all. This paper reviews two statistical techniques for classification of events in order to identify which should trigger bond payments based on a large sample of simulated earthquakes. These statistical techniques are effective, simple to interpret and to implement. A numerical experiment is performed to illustrate their use, and to facilitate a comparison with a previously published evolutionary computation algorithm.

A simheuristic algorithm to set up starting times in the stochastic parallel flowshop problem

Abstract This paper addresses the parallel flowshop scheduling problem with stochastic processing times, where a product composed of several components has to be finished at a particular moment. These components are processed in independent parallel factories, and each factory can be modeled as a permutation flowshop. The processing time of each operation at each factory is a random variable following a given probability distribution. The aim is to find the robust starting time of the operations at each factory in a way that all the components of the product are completed on a given deadline with a user-defined probability. A simheuristic algorithm is proposed in order to minimize each of the following key performance indicators: (i) the makespan in the deterministic version; and (ii) the expected makespan or a makespan percentile in the stochastic version. A set of computational experiments are carried out to illustrate the performance of the proposed methodology by comparing the outputs under different levels of stochasticity.

Learnheuristics: hybridizing metaheuristics with machine learning for optimization with dynamic inputs

Abstract This paper reviews the existing literature on the combination of metaheuristics with machine learning methods and then introduces the concept of learnheuristics, a novel type of hybrid algorithms. Learnheuristics can be used to solve combinatorial optimization problems with dynamic inputs (COPDIs). In these COPDIs, the problem inputs (elements either located in the objective function or in the constraints set) are not fixed in advance as usual. On the contrary, they might vary in a predictable (non-random) way as the solution is partially built according to some heuristic-based iterative process. For instance, a consumer’s willingness to spend on a specific product might change as the availability of this product decreases and its price rises. Thus, these inputs might take different values depending on the current solution configuration. These variations in the inputs might require from a coordination between the learning mechanism and the metaheuristic algorithm: at each iteration, the learning method updates the inputs model used by the metaheuristic.

Combining simulation with metaheuristics in distributed scheduling problems with stochastic processing times

In this paper, we focus on a scenario in which a company or a set of companies conforming a supply network must deliver a complex product (service) composed of several components (tasks) to be processed on a set of parallel flow-shops with a common deadline. Each flow-shop represents the manufacturing of an independent component of the product, or the set of activities of the service. We assume that the processing times are random variables following a given probability distribution. In this scenario, the product (service) is required to be finished by the deadline with a user-specified probability, and the decision-maker must decide about the starting times of each component/task while minimizing one of the following alternative goals: (a) the maximum completion time; or (b) the accumulated deviations with respect to the deadline. A simheuristic-based methodology is proposed for solving this problem, and a series of computational experiments are performed.

Solving Realistic Portfolio Optimization Problems via Metaheuristics: A Survey and an Example

Computational finance has become one of the emerging application fields of metaheuristic algorithms. In particular, these optimization methods are quickly becoming the solving approach alternative when dealing with realistic versions of financial problems, such as the popular portfolio optimization problem (POP). This paper reviews the scientific literature on the use of metaheuristics for solving rich versions of the POP and illustrates, with a numerical example, the capacity of these methods to provide high-quality solutions to complex POPs in short computing times, which might be a desirable property of solving methods that support real-time decision making.

Statistical and machine learning approaches for the minimization of trigger errors in parametric earthquake catastrophe bonds

Catastrophe bonds are financial instruments designed to transfer risk of monetary losses arising from earthquakes, hurricanes, or floods to the capital markets. The insurance and reinsurance industry, governments, and private entities employ them frequently to obtain coverage. Parametric catastrophe bonds base their payments on physical features. For instance, given parameters such as magnitude of the earthquake and the location of its epicentre, the bond may pay a fixed amount or not pay at all. This paper reviews statistical and machine learning techniques for designing trigger mechanisms and includes a computational experiment. Several lines of future research are discussed.

A SimILS-Based Methodology for a Portfolio Optimization Problem with Stochastic Returns

Combinatorial optimization has been a workhorse of financial and risk management, and it has spawned a large number of real-life applications. Prominent in this body of research is the mean-variance efficient frontier (MVEF) that emanates from the portfolio optimization problem (POP), pioneered by Harry Markowitz. A textbook version of POP minimizes risk for a given expected return on a portfolio of assets by setting the proportions of those assets. Most authors deal with the variability of returns by employing expected values. In contrast, we propose a simILS-based methodology (i.e., one extending the Iterated Local Search metaheuristic by integrating simulation), in which returns are modeled as random variables following specific probability distributions. Underlying simILS is the notion that the best solution for a scenario with expected values may have poor performance in a dynamic world.