Maritime operations (naval refueling, amphibious landing, etc.) are heavily affected by meteorological and oceanographic conditions (METOC) on short and medium time/range scales, both in ports (shallow waters) and at sea/ocean (deep waters). Exploiting METOC conditions to optimally plan the deployment position and/or the paths of the assets involved in a maritime mission (underwater vehicles, surface vehicles, etc.), is a challenging problem, frequently involving more than a single objective to optimize. Socio/economical/political factors driving the initial planning of operations, for example, or the amount of vessel traffic or the fishing operations through the areas of interest, must be taken into account in such situations. Multiobjective Evolutionary Algorithms (MEAs) are proving to be a powerful tool to solve such problems. After reviewing the state-of-the-art in multiobjective evolutionary optimization, we discuss the current challenges when using MEAs in asset planning: the number of objectives is high, the objectives can be noisy, and the fitness evaluation can be very computationally complex. We will address the last issue with particular attention, showing how to parallelize MEAs, how to exploit General Purpose Graphics Processing Units (GPGPU), how to approximate the fitness, and how to use surrogate models.
Challenges in Multi-Objective Evolutionary Optimization for Asset Planning Decision Support
COCOCCIONI, MARCO
2015-01-01
Abstract
Maritime operations (naval refueling, amphibious landing, etc.) are heavily affected by meteorological and oceanographic conditions (METOC) on short and medium time/range scales, both in ports (shallow waters) and at sea/ocean (deep waters). Exploiting METOC conditions to optimally plan the deployment position and/or the paths of the assets involved in a maritime mission (underwater vehicles, surface vehicles, etc.), is a challenging problem, frequently involving more than a single objective to optimize. Socio/economical/political factors driving the initial planning of operations, for example, or the amount of vessel traffic or the fishing operations through the areas of interest, must be taken into account in such situations. Multiobjective Evolutionary Algorithms (MEAs) are proving to be a powerful tool to solve such problems. After reviewing the state-of-the-art in multiobjective evolutionary optimization, we discuss the current challenges when using MEAs in asset planning: the number of objectives is high, the objectives can be noisy, and the fitness evaluation can be very computationally complex. We will address the last issue with particular attention, showing how to parallelize MEAs, how to exploit General Purpose Graphics Processing Units (GPGPU), how to approximate the fitness, and how to use surrogate models.File | Dimensione | Formato | |
---|---|---|---|
DeSRAAP_2015_abstract.pdf
solo utenti autorizzati
Descrizione: Abstract
Tipologia:
Abstract
Licenza:
NON PUBBLICO - Accesso privato/ristretto
Dimensione
139.62 kB
Formato
Adobe PDF
|
139.62 kB | Adobe PDF | Visualizza/Apri Richiedi una copia |
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.