Surrogate Model-based Multi-Objective Optimization in Early Stages of
Ship Design
Dublin Core
Title
Surrogate Model-based Multi-Objective Optimization in Early Stages of
Ship Design
Ship Design
Subject
ship design, multi-objective optimization, surrogate model, neural network, particle swarm optimizer
Description
The abstract isthe early stages of ship design, the decision of the ship's main dimensions significantly impacts the ship's
performance and the total cost of ownership. This paper focuses on an optimization approach based on surrogate models at
the early stages of ship design. The objectives are to minimize power requirements and building costs while still satisfying the
constraints. We compare three approaches of surrogate models: Kriging, BPNN-PSO (Backpropagation Neural NetworkParticle Swarm Optimizer), and MLP (Multi-Layer Perceptron) in two multi-objective optimization algorithms: MOEA/D
(Multi-Objective Evolutionary Algorithm Decomposition) and NSGA-II (Non-Dominated Sorting Genetic Algorithm II). The
experimental results show that MLP surrogate models get the best performance with MAE 6.03, and BPNN-PSO gets the
second position with MAE 7.2. BPNN-PSO and MLP with MOEA/D and NSGA-II improve the design with around 58% smaller
adequate power and 6% less steel weight than the original design. However, BPNN-PSO and MLP have lower hypervolume
than Kriging for both optimization algorithms MOEA/D and NSGA-II. On the other hand, Kriging has the most inadequate
model accuracy performance, with an MAE of 22.2, but produces the highest hypervolume, lowest computational time, and far
lower objective values than BPNN-PSO and MLP for both optimization algorithms, MOEA/D and NSGA-II. Nevertheless, the
three surrogate model approaches can significantly improve ship design solutions and reduce work time in the early stages of
design
performance and the total cost of ownership. This paper focuses on an optimization approach based on surrogate models at
the early stages of ship design. The objectives are to minimize power requirements and building costs while still satisfying the
constraints. We compare three approaches of surrogate models: Kriging, BPNN-PSO (Backpropagation Neural NetworkParticle Swarm Optimizer), and MLP (Multi-Layer Perceptron) in two multi-objective optimization algorithms: MOEA/D
(Multi-Objective Evolutionary Algorithm Decomposition) and NSGA-II (Non-Dominated Sorting Genetic Algorithm II). The
experimental results show that MLP surrogate models get the best performance with MAE 6.03, and BPNN-PSO gets the
second position with MAE 7.2. BPNN-PSO and MLP with MOEA/D and NSGA-II improve the design with around 58% smaller
adequate power and 6% less steel weight than the original design. However, BPNN-PSO and MLP have lower hypervolume
than Kriging for both optimization algorithms MOEA/D and NSGA-II. On the other hand, Kriging has the most inadequate
model accuracy performance, with an MAE of 22.2, but produces the highest hypervolume, lowest computational time, and far
lower objective values than BPNN-PSO and MLP for both optimization algorithms, MOEA/D and NSGA-II. Nevertheless, the
three surrogate model approaches can significantly improve ship design solutions and reduce work time in the early stages of
design
Creator
Nanda Yustina1
, Ari Saptawijaya2
, Ari Saptawijaya2
Publisher
Universitas Indonesia,
Date
31-10-2022
Contributor
Fajar bagus W
Format
PDF
Language
Indonesia
Type
Text
Files
Collection
Citation
Nanda Yustina1
, Ari Saptawijaya2, “Surrogate Model-based Multi-Objective Optimization in Early Stages of
Ship Design,” Repository Horizon University Indonesia, accessed June 7, 2025, https://repository.horizon.ac.id/items/show/9238.
Ship Design,” Repository Horizon University Indonesia, accessed June 7, 2025, https://repository.horizon.ac.id/items/show/9238.