Emergent optimizationdesign and applications in telecommunications and bioinformatics

Resumen

In this PhD Thesis, we are interested in designing new Particle Swarm Optimization (PSO) proposals that solve or mitigate the main disadvantages present in this algorithm. To this point, we have used a series of methodologies to analyze the internal behavior of this algorithm and to identify the main problems or improvement opportunities appearing in existing PSO versions. In order to assess the effectiveness of the new proposals, we have performed comparative studies from two main points of view: solution quality and scalability in terms of the problem size (decision variables). For this task we have followed specific experimental procedures of standard benchmark test suites (CEC05, SOCO10, DTLZ, etc.), and we have compared against the most prominent metaheuristics in current state of the art (G-CMA-ES, MSGA-II, OMOPSO, DE, MOS-DE, and MTS). In this context we have proposed a series of new algorithmic approaches: DEPSO, RPSO-vm, PSO6, SMPSO, PSO6-Mtsls, that have been validated and located at the top level outperforming algorithms in the state of the art. Second, we are aimed at solving real world complex problems with PSO based algorithms to determine the adaptability of this algorithm to different representations and scenario conditions, within limited computational time, and requiring huge data base management. In concrete, we have focused in this thesis on three NP-Hard real applications trying to cover quite different industry fields: Gene Selection in DNA Microarrays (Bioinformatics), Communication Protocol Tuning in VANETs (Telecommunications), and Signal Lights Timing in traffic management (Urban Mobility).