.png){kind=icon}
.png){kind=icon}
Return
Diversity metrics for direct-coded variable-length chromosome shortest path problem evolutionary algorithms
.png){kind=icon}
PRE
AI
DOI：10.1007/s00607-020-00851-4.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
One of the major concerns in Evolutionary Algorithms is the premature convergence caused by what is known as the Exploration and Exploitation Balance problem. To maintain this balance, population diversity should be maintained during the initialization/optimization process. Maintaining diversity can be done through different strategies, but they commonly answer one question: when to introduce more diversity to the population? To answer this question there should be diversity metrics upon which a decision can be made to add diversity; consequently, add/reduce exploration/exploitation. There are as many diversity metrics as many problems and representations. That is, diversity metrics are very problem-specific. This work provides diversity metrics for the variable-length chromosome Genetic Algorithm for Shortest Path. The suggested metrics consider the varying lengths of the chromosomes, problem representation, and the search space. To measure chromosome-length diversity, a novel chromosome-length-based metric has been proposed. By exploiting the fact that the possible genes that can form any chromosome are well known in this specific direct-encoded population, a new simple metric that measures the representation of genes in the initial population is proposed and experimentally investigated. The presented metrics put through an extensive simulation and comparatively studied. Relationship between the proposed metrics has been quantified using Principal Component Analysis under varying network/population sizes.
Keywords:
Shortest path problem
Diversity
Genetic algorithm
Evolutionary algorithms
AI Summary
Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.