Carolina Salto

Enhanced evolutionary algorithms for single and multiobjective optimization in the job shop scheduling problem

Abstract Over the past few years, a continually increasing number of research efforts have investigated the application of evolutionary computation techniques for the solution of scheduling problems. Scheduling can pose extremely complex combinatorial optimization problems, which belong to the NP-hard family. Last enhancements on evolutionary algorithms include new multirecombinative approaches. Multiple Crossovers Per Couple (MCPC) allows multiple crossovers on the couple selected for mating and Multiple Crossovers on Multiple Parents (MCMP) do this but on a set of more than two parents. Techniques for preventing incest also help to avoid premature convergence. Issues on representation and operators influence efficiency and efficacy of the algorithm. The present paper shows how enhanced evolutionary approaches, can solve the Job Shop Scheduling Problem (JSSP) in single and multiobjective optimization.

Designing heterogeneous distributed GAs by efficiently self-adapting the migration period

This paper investigates a new heterogeneous method that dynamically sets the migration period of a distributed Genetic Algorithm (dGA). Each island GA of this multipopulation technique self-adapts the period for exchanging information with the other islands regarding the local evolution process. Thus, the different islands can develop different migration settings behaving like a heterogeneous dGA. The proposed algorithm is tested on a large set of instances of the Max-Cut problem, and it can be easily applied to other optimization problems. The results of this heterogeneous dGA are competitive with the best existing algorithms, with the added advantage of avoiding time-consuming preliminary tests for tuning the algorithm.

Analysis of distributed genetic algorithms for solving cutting problems

In this paper, a solution to the three-stage two-dimensional cutting problem is presented by using sequential and parallel genetic algorithms (GAs). More specifically, an analysis of including distributed population ideas and parallelism in the basic GA are carried out to solve the problem more accurately and efficiently than with ordinary sequential techniques. Publicly available test problems have been used to illustrate the computational performance of the resulting metaheuristics. Experimental evidence in this work will show that the proposed algorithms outperform their sequential counterparts in time (high speedup with multiprocessors) and numerically (lower number of visited points during the search to find the solutions).

Comparison of Recombination Operators in Panmictic and Cellular GAs to Solve a Vehicle Routing Problem

The Vehicle Routing Problem (VRP) deals with the assignment of a set of transportation orders to a fleet of vehicles, and the sequencing of stops for each vehicle to minimize transportation costs. In this paper we study the Capacitated VRP (CVRP), which is mainly characterized by using vehicles of the same capacity. Taking a basic GA to solve the CVRP, we propose a new problem dependent recombination operator, called Best Route Better Adjustment recombination (BRBAX). A comparison of its performance is carried out with respect to other two classical recombination operators. Also we conduct a study of different mutations in order to determine the best combination of genetic operators for this problem. The results show that the use of our specialized BRBAX recombination outperforms the others more generic on all problem instances used in this work for all the metrics tested.

Using landscape measures for the online tuning of heterogeneous distributed gas

Tuning distributed genetic algorithms (dGAs) increases even more the task of finding an appropriate parameterization, since the migration operator adds, at least, five additional values that have to be set up. This work is a preliminary approach on using a landscape measure (the Fitness Distance Correlation) to dynamically adjust one of these five parameters, in particular, the migration period. The results have shown that, by using this information, the quality of the solutions is competitive with those obtained by the algorithms with the pre-tuned migration period, but with a saving of more than 100 hours of preliminary experimentation.

Hybrid Ant Colony System to Solve a 2-Dimensional Strip Packing Problem

In this paper we present a study of an Ant Colony System (ACS) for the two-dimensional strip packing problem. In our computational study, we emphasize the influence of incorporating a simple optimization method at each cycle of the ACS. In this hybrid approach, local optimization is applied to a subset of the newly generated solutions to move them to a local optimum. We show that our ACS algorithm, when combined with a fine-tuned local search procedure, can compete with an existing genetic algorithm, reaching solutions of good quality and also exhibiting low execution times.

Heterogeneity through Proactivity: Enhancing Distributed EAs

This work proposes an heterogeneous distributed evolutionary algorithm that automatically adapts its migration policy based on the entropy of the population. It is an heterogeneous algorithm since the search performed by each subpopulation is different from each other. the novelty of our approach lies on its proactivity, in which each subpopulation can ask for more/less frequent migrations from its neighbors in order to maintain the genetic diversity at a desired level. the goal is to avoid the subpopulations to get trapped into local minima. the results on large NK-landscape instances have shown that the proactive strategy is a very promising approach, specially for highly rugged landscapes, in which it does not only reaches the most accurate solutions, but it does the fastest.

Enhancing distributed EAs by a proactive strategy

In this work we propose a new distributed evolutionary algorithm that uses a proactive strategy to adapt its migration policy and the mutation rate. The proactive decision is carried out locally in each subpopulation based on the entropy of that subpopulation. In that way, each subpopulation can change their own incoming flow of individuals by asking their neighbors for more frequent or less frequent migrations in order to maintain the genetic diversity at a desired level. Moreover, this proactive strategy is reinforced by adapting the mutation rate while the algorithm is searching for the problem solution. All these strategies avoid the subpopulations to get trapped into local minima. We conduct computational experiments on large instances of the NK landscape problem which have shown that our proactive approach outperforms traditional dEAs, particularly for not highly rugged landscapes, in which it does not only reaches the most accurate solutions, but it does the fastest.

Enhancing distributed EAs using proactivity

In this abstract we describe a proactive strategy followed by a distributed evolutionary algorithm to adapt its migration policy. The proactive decision is made locally within each subpopulation, ant it is based on the entropy of that subpopulation. In that way, each subpopulation can ask for more/less frequent migrations from its neighbors in order to maintain the genetic diversity at a desired level, thus avoiding the subpopulations to get trapped into local minima. We conduct computational experiments on a set of different problems and it is shown that our proactive approach outperforms classical dEA settings by reaching accurate solutions in a lower number of generations.

Distributed Evolutionary Algorithms in Heterogeneous Environments

Distributed computing environments are usually composed of many heterogeneous computers able to work cooperatively. We analyze the impact in the performance of a parallel metaheuristic when it is executed using a set of heterogeneous computing resources. Following a well-defined methodology, the aim of the paper is to use all the computing resources but at the same time to be efficient in time. Our conclusion is that both the solution quality and the numeric effort are comparable to that achieved by using a (faster) homogeneous platform, the traditional environment to execute this kind of algorithms.