A Parallel Plugin-Based Framework for Multi-objective Optimization

摘要

This work presents a parallel framework for the solution of multi-objectiveoptimization problems. The framework implements some of the best known multi-objective evolutionary algorithms. The framework architecture makes usage of config-uration files to provide a more extensive and simple customization environment thanother similar tools. A wide variety of configuration options can be specified to adaptthe software behaviour to many different parallel models, including a new adaptivemodel which dynamically grants more computational resources to the most promisingalgorithms. The plugin-based architecture of the framework minimizes the final usereffort required to incorporate their own problems and evolutionary algorithms, andfacilitates the tool maintenance. The flexibility of the approach has been tested byconfiguring a standard homogeneous island-based model and a self-adaptive model.The computational results obtained for problems with different granularity demons-trate the efficiency of the provided parallel implementation.