Abstract
Most real-world optimization problems feature several (possibly conicting) objectives that must be simultaneously satised. In words, they are multi-objective in nature. Consequently, the notion of optimality is to be redened in such a way that instead of aiming at nding a single solution, a set of compromises or trade-os are pursued upon whose completion the desired solution is selected. Despite the considerable increase in the number of novel meta-heuristic algorithms for solving optimization problems in the past decades, such techniques have not been extrapolated to multi-objective problem formulations at the same pace, as less research eorts have been conducted towards this line. Therefore, the adaptation of existing single-objective algorithms for simultaneously dealing with several objectives,as well as the development of advanced multi-objective approaches are deemed crucial for handling the current requirements of real-world optimization problems. In this context, the Harmony Search algorithm (HS), a single-objective optimization solver developed in the 2000's, has been widely proven to obtain excellent results in the eld of combinatorial optimization. Its innovative features and probabilistic parameters' characteristics makes Harmony Search outperform other meta-heuristic algorithms in the literature, thereby promoting its utilization in application elds arising from
diverse knowledge areas, such as construction engineering, telecommunications and economics.
This being exposed, this Thesis sheds light on the adaptation of the Harmony
Search heuristic to multi-objective optimization paradigms, and the applicability of the resulting novel multi-objective approach (coined as Non-Dominated Sorting Harmony Search, NSHS-II) to dierent complex optimization problems. Specically, four are the application scenarios where NSHS-II is put to practice: 1) localization in wireless sensor networks; 2) reconguration of urban road networks; 3) design of WiFi access networks; and 4) distribution of healthcare resources. Since the denition of the best encoding strategy for each problem is of utmost importance to univocally represent each encountered solution and hence, avoid possible redundancy, grouping encoding strategies are also investigated in this dissertation on the purpose of eliminating inherently redundant solutions and reduce the computational complexity
of the search process. The proposed NSHS-II approach is validated by means of
exhaustive experimental benchmarks to algorithmic counterparts from the state of the art, which verify that the NSHS-II algorithm is a cross-eld robust technique capable of outperforming other existing multi-objective approaches in multi-objective optimization paradigms.
| Date of Award | 2013 |
|---|---|
| Original language | English |
| Awarding Institution |
|