Catálogo de publicaciones - libros

Since genetic algorithms (GAs) treat a population of finite size, it is necessary to study stochastic fluctuations in evolution processes. In this study, we investigated the influence of genetic drift due to finite population size on the performance of a GA on the multiplicative landscape. There was large difference between numerical experiments with small population size and the prediction of deterministic model. It was observed in some experiments that favorable first order schemata were lost from the population. It was also noted that the population can be assumed to be in linkage equilibrium in the GA including crossover. Then we performed the theoretical investigation of frequencies of the first order schemata, and calculated their changes in time by using the Wright-Fisher model and diffusion equations. We showed that these mathematical theories reasonably predict various quantities including the ultimate extinction probability. We found that the extinction of favorable schemata is the most undesirable effect of genetic drift.

Memetic evolutionary algorithms (MEAs) combine the global search of evolutionary learning methods and the fine-tune ability of local search methods so that they are orders of magnitude more accurate than traditional evolutionary algorithms in many problem domains. However, little work has been done on the mathematical model and convergence analysis of MEAs. In this paper, a theoretical model as well as the convergence analysis of a class of gradient-based MEAs is presented. The results of this paper are extensions of the research work on the abstract model and convergence analysis of general evolutionary algorithms. By modeling the local search of gradient methods as an abstract strong evolution operator, the theoretical framework for abstract memetic evolutionary algorithms is derived. Moreover, the global convergence theorems and the convergence rate estimations of gradient-based MEAs are also established.

In the case of multiobjective evolutionary algorithm, the outcome is usually an approximation of the true Pareto Optimal set and how to evaluate the quality of the approximation of the Pareto-optimal set is very important. In this paper, improved measures are carried out to the approximation, uniformity and well extended for the approximation of the Pareto optimal set with the advantage of easy to operate. Finally, we apply our measures to the four multiobjective evolutionary algorithms that are representative of the state-of-the-art on the standard functions. Results indicate that the measures are highly competitive and can be conducted to the comparisons of the approximation set.

A new dynamic evolutionary algorithm based on orthogonal design (denoted by ) is proposed in present paper. Its population does not consist of individuals (solution vectors), but of niches, a properly small hyper-rectangle where orthogonal design method likely work well. Each niche selects the best solution found so far as its representative. And orthogonal design method is employed to find potentially good solution which is probably the representative in the niche. The niche mutation, the only genetic operator in this evolutionary algorithm, is guided by the representative of the niche, therefore, the fitness of the offspring is likely better than that of its father, furthermore, evolves fast. We employ a complex benchmark (moving peaks functions) testing the new approach and the numerical experiments show that performs much better than [1].

Fitness sharing genetic algorithm is one of the most common used methods to deal with multimodal optimization problems. The algorithm requires peaks radii as the predefined parameter. It is very difficult to guess peaks radii for designers and decision makers in the real world applications. A novel self-adaptive annealing peaks radii control method has been suggested in this paper to deal with the problem. Peaks radii are coded into chromosomes and evolved while fitness sharing genetic algorithm optimizes the problem. The empirical results tested on the benchmark problems show that fitness sharing genetic algorithm with self-adaptive annealing peaks radii control method can find and maintain nearly all peaks steadily. This method is especially suitable for the problems whose peaks radii are difficult to estimate beforehand.

The hybrid multimodal optimization algorithm that combines a novel clustering method and fitness sharing method is presented in this paper. The only parameter required by the novel clustering method is the peak number. The clustering criteria include minimizing the square sum of the inner-group distance, maximizing the square sum of the inter-group distance, and the fitness value of the individuals. After each individual has been classified to the certain cluster, fitness sharing genetic algorithm is used to find multiple peaks simultaneously. The empirical study of the benchmark problems shows that the proposed method has satisfactory performance.

The differential evolution (DE) is a stochastic, population-based, and relatively unknown evolutionary algorithm for global optimization that has recently been successfully applied to many optimization problems. This paper presents a new variation on the DE algorithm, called the cooperative co-evolutionary differential evolution (CCDE). CCDE adopts the cooperative co-evolutionary architecture, which was proposed by Potter and had been successfully applied to genetic algorithm, to improve significantly the performance of the DE. Such improvement is achieved by partitioning a high-dimensional search space by splitting the solution vectors of DE into smaller vectors, then using multiple cooperating subpopulations (or smaller vectors) to co-evolve subcomponents of a solution. Applying the new DE algorithm to on 11 benchmark functions, we show that CCDE has a marked improvement in performance over the traditional DE and cooperative co-evolutionary genetic algorithm (CCGA).

Optimal design for urban mass transit network is the precondition and basis to establish an effective public transportation system. Transit network optimization and headway optimization are two of the most important issues to be dealt with. In this paper, a transit network optimization model is firstly proposed to maximize the nonstop passenger flow. Moreover, this paper puts forwards the optimization model of headways for all the transit routes in the optimized network. Since all the two models can be boiled down to the NP-hard problem, two kinds of evolutionary algorithms, i.e., ant colony algorithm and improved genetic algorithm are introduced to solve the problems respectively. Finally, a case study in a typical city is introduced to explain the validity of the proposed methods.

This paper develops a novel method for solving a type of nonlinear programming model with all fuzzy coefficients (AFCNP). For a decision maker specified credibility level, by presenting the equivalent deterministic forms of fuzzy inequality constraints and fuzzy objective, the fuzzy model is converted into a crisp constrained nonlinear programming model with parameter (CPNP). An improved genetic algorithm is presented to solve the CPNP and obtain the crisp optimal solution of AFCNP for specified credibility level.

In this paper, we propose an evolutionary algorithm based on a single operator called stochastic weighted learning, i.e., each individual will learn from other individuals specified with stochastic weight coefficients in each generation, for constrained optimization. For handling equality and inequality constraints, the proposed algorithm introduces a learning rate adapting technique combined with a fitness comparison schema. Experiment results on a set of benchmark problems show the efficiency of the algorithm.