Thesis Diseñar una librería de métricas de desempeño para metaheuristícas poblacionales
Loading...
Date
2024
Authors
Journal Title
Journal ISSN
Volume Title
Program
Ingeniería Civil Informática
Departament
Campus
Campus Santiago San Joaquín
Abstract
Las metaheurísticas son algoritmos que resuelven problemas complejos en tiempos limitados, ofreciendo soluciones de buena calidad, aunque sin garantizar el óptimo. Para medir su desempeño, se suelen usar métricas simples como la aptitud de las soluciones y el tiempo de ejecución. Usualmente, estas métricas simples no requieren de mayor desarrollo ni procesamiento, pero que carecen de análisis profundo sobre el algoritmo. Este trabajo presenta una librería que implementa métricas de desempeño para metaheurísticas poblacionales, analizando su comportamiento en sus etapas de exploración y explotación de la solución. Para evaluar la propuesta, se utilizaron tres algoritmos poblacionales: un algoritmo genético estándar (vGA) y dos variantes de un algoritmo genético rápido (fGA), las cuales fueron usadas en un análisis comparativo. Los experimentos validan estas métricas, la información que entregan y su interpretación para evaluar o comparar algoritmos a través de gráficos con datos históricos de la ejecución del algoritmo.
Metaheuristics are algorithms that solve complex problems within limited timeframes, offering high-quality solutions, although without guaranteeing optimality. To measure their performance, simple metrics such as solution quality and execution time are often used. Usually, these metrics require little development or processing but lack in-depth analysis of the algorithm. This work presents a library that implements performance metrics for population-based metaheuristics, analyzing their behavior in the exploration and exploitation stages of the solution. To evaluate the proposal, three population-based algorithms were used: a standard genetic algorithm (vGA) and two variants of a fast genetic algorithm (fGA), which were used in a comparative analysis. The experiments validate these metrics, the information they provide, and their interpretation for evaluating or comparing algorithms through graphs with historical data from the algorithm’s execution.
Metaheuristics are algorithms that solve complex problems within limited timeframes, offering high-quality solutions, although without guaranteeing optimality. To measure their performance, simple metrics such as solution quality and execution time are often used. Usually, these metrics require little development or processing but lack in-depth analysis of the algorithm. This work presents a library that implements performance metrics for population-based metaheuristics, analyzing their behavior in the exploration and exploitation stages of the solution. To evaluate the proposal, three population-based algorithms were used: a standard genetic algorithm (vGA) and two variants of a fast genetic algorithm (fGA), which were used in a comparative analysis. The experiments validate these metrics, the information they provide, and their interpretation for evaluating or comparing algorithms through graphs with historical data from the algorithm’s execution.
Description
Keywords
Métrica de rendimiento, Optimización, Metaheurística poblacional