Browsing by Subject "Genetic Algorithm"

Now showing 1 - 3 of 3

Open Access
Approximating small open economy models with neural network trained by genetic algorithm
(2008) Coşkun, Yeşim
This thesis work presents a direct numerical solution methodology to approximate the small open economy models with debt elastic interest rate premium and with convex portfolio adjustment cost, both studied by Stephanie SchmittGroh´e and Martin Uribe(2003). This recent method is compared with the firstorder approximation to the policy function from the aspect of second moments of endogenous variables and their impulse responses. The proposed methodology, namely genetic algorithm-neural network (GA-NN), parameterizes the policy function with a feed-forward neural network that is trained by a genetic algorithm. Thus, unlike the first-order approximation, GA-NN does not require the continuity and the existence of derivatives of objective and policy functions. Importantly, since genetic algorithm is an evolutionary algorithm that enables global search over the feasible set, it provides a robust result in any solution space. Also GA-NN method gives not only the moments of the model but also the optimal path.
Open Access
Parameter selection for genetic algorithm-based simulation optimization
(2001) Boyabatlı, Onur
Improvements on heuristic techniques with the availability of faster PC’s increase the importance of simulation-optimization (sim/opt) applications. Sim/opt methodologies use computer simulation integrated with an optimization sub-routine to optimize the problems of interest. The main contribution of these methods is to make simulation as a prescriptive tool rather than a descriptive tool, which has been widely used as the descriptive tool for estimating the performance of complex stochastic systems. Sim/opt methodologies have been applied on various combinatorial optimization problems, and the current trend in sim/opt area is the use of meta-heuristic techniques. Genetic Algorithm (GA) is the well known metaheuristic, which is a global search algorithm taking its inspiration from natural genetics. GA has several parameters affecting its performance. Even for the GA with same structural parameters (coding scheme, operator types, stopping criterion), the different combinations of numerical parameters (initial population type, population size, maximum generation number and the crossover and mutation probabilities) may lead to drastic changes in the performance of the algorithm. This study examines the effects of numerical parameters of GA on its performance in terms of both fitness and CPU time; and proposes guidelines for appropriate parameter selection. A test problem of a serial assembly line taken from the literature is used for the GA-based simulation-optimization application. A genetic algorithm coded in C is integrated with a simulation model developed using SIMAN simulation language. Modifications on the test problem are made to analyze the behavior of GA parameters under different experimental conditions. The computational results reveal that in the case of a dominant set of decision variables, for rapid convergent GA applications high mutation rates are more useful, whereas the crossover operator does not have any significant impact on GA performance.
Open Access
Parameter selection in genetic algorithms
(International Institute of Informatics and Cybernetics, 2004) Boyabatlı, O.; Sabuncuoglu, I.
In this study, we provide a new taxonomy of parameters of genetic algorithms (GA), structural and numerical parameters, and analyze the effect of numerical parameters on the performance of GA based simulation optimization applications with experimental design techniques. Appropriate levels of each parameter are proposed for a particular problem domain. Controversial to existing literature on GA, our computational results reveal that in the case of a dominant set of decision variable the crossover operator does not have a significant impact on the performance measures, whereas high mutation rates are more suitable for GA applications.