Parallel processing for dynamic multi-objective optimization

Resumen

The main objective of this PhD thesis is to advance the field of parallel multi-objective evolutionary algorithms to solve dynamic multi-objective optimization problems, Thus, the research presented in this thesis involves three different, although related, fields: - Multi-objective evolutionary algorithms (MOEA), - Dynamic multi-objective optimization (DMO) problems, and - Parallelization of MOEAs to solve DMO problems. The degree of advancement of the research varies for each of the afore-mentioned topics, from a full-fledged research field as it is the MOEA topic to a new emerging subject as it happens with dynamic multi-objective optimization. Nevertheless, proposals to improve further the three afore-mentioned subjects have been made in this thesis. First of all, this thesis introduces a \textit{low-cost} MOEA able to deal with multi-objective problems within more restrictive time limits than other state-of-the-art can do. Secondly, the field of dynamic optimization is reviewed and some additions are made so that the field moves forward to tackle dynamic multi-objective problems. This has been facilitated by the introduction of performance measures for problems that are both dynamic and multi-objective. Moreover, modifications are proposed for two of the five \textit{de facto} standard test cases for DMO problems. Thirdly, the parallelization of MOEAs to solve DMO problems is addressed with two different proposed approaches: - A hybrid master-worker and island approach called pdMOEA, and - A fully distributed approach called pdMOEA+. These two approaches are compared side-by-side with the test cases already mentioned. Finally, future work to follow upon the achievements of this thesis is outlined.