1 INTRODUCTION

Maritime transport is the backbone of global logistics,

is heavily relied upon to ship goods between countries

and regions[1]. It also enhances national

competitiveness by integrating countries into global

trade networks. In recent years, trade and logistics

exchanges have grown. According to the International

Maritime Organization (IMO), around 11 billion tons

of goods are transported by sea annually, accounting

for over 80% of world trade by volume and more than

70% by value[2]. This makes it the dominant mode of

Towards Improved Ship Weather Routing Through

Multi-Objective Optimization with High Performance

Computing Support

M. Abdalsalam & J. Szłapczyńska

Gdansk University of Technology, Gdańsk, Poland

ABSTRACT: Maritime transport remains integral to the global economy, facilitating the cost-efficient and scalable

movement of cargo and individuals over varying distances. Modern and effective ship routing solutions not only

minimize voyage time and operational costs (including fuel consumption), but also improve resource allocation

and environmental sustainability. Their planning process relies heavily on optimization algorithms capable of

addressing numerous environmental and operational constraints, particularly in the context of dynamic and often

unpredictable weather conditions.

A widely adopted approach in the literature is to formulate the ship route optimization problem as a multi-

objective optimization (MOO) task, incorporating both static and dynamic constraints. The complexity of this

formulation increases significantly when uncertainties related to weather conditions and ship behaviour are

introduced, further complicating the optimization process. Meta-heuristic algorithms have gained prominence as

effective tools for addressing i.e. complex multi-objective, constrained and nonlinear problems. Despite their

demonstrated computational efficiency, the overall process of ship weather route optimization often remains

computationally intensive, posing significant challenges for real-time or near-real-time applications in

operational maritime contexts.

High Performance Computing (HPC) emerges as a viable approach to overcome this limitation. HPC refers

mostly to the use of advanced computational systems composed of parallel processing architectures (such as

CUDA, OpenMP, MPI, among others) to solve much faster and more efficiently complex and data-intensive

problems. Originating in scientific research domains, HPC technologies have rapidly evolved and are now being

applied to support solving a wide range of problem in computer science and engineering. Employing HPC-

enabled computations allows for designing a scalable and efficient framework for tackling the growing

complexity of ship weather routing.

This paper discusses the possibilities of HPC integration with MOO for ship weather routing, aiming to

demonstrate how HPC-enabled methodologies can improve the performance and real-life applicability of the

routing systems.

http://www.transnav.eu

the International Journal

on Marine Navigation

and Safety of Sea Transportation

Volume 19

Number 1

March 2025

DOI: 10.12716/1001.19.01.12

international trade due to its superior capacity to

handle large quantities of goods at low costs[3].

However, it faces various challenges that affect its

efficiency and effectiveness. A key issue is the rising

cost of fuel due to global price fluctuations, which

increases operating expenses. The sector also faces

growing environmental pressures, including

significant demands to reduce carbon emissions

doubts regarding alternative fuel options. Maritime

shipping accounts for approximately 3% of global

carbon dioxide emissions[1]. As shown in Figure 1,

since 1970, shipping emissions have more than

doubled, rising from about 350 million to over 700

million metric tons by 2023[4]. According to Statista,

the shipping sector accounted for 12% of transport-

related CO₂ emissions in 2023, with 4.5% from

domestic and 7.5% from international shipping[5].

The situation necessitates a shift toward more

sustainable modes and technologies to reduce negative

environmental impacts. On the other hand, climate

change and extreme weather disrupted shipping

operations, affecting global supply chains and

increasing accident risks.

(a)

(b)

Figure 1. Environmental impact of shipping [4]. a) Carbon

dioxide emissions from international shipping worldwide

(1970–2023, million metric tons); b) Distribution of carbon

dioxide emissions by transportation subsector worldwide

(2023)

These factors highlight the need for effective

measures to ensure safety and minimize weather-

related disruptions. Considering these growing

environmental and climate challenges, the need for

innovative solutions that address emissions and safety

challenges while enhancing efficiency and reducing

costs has emerged. Among the most prominent of these

solutions is the development of advanced vessel

routing systems that take weather conditions into

account, simultaneously improving efficiency and

reducing fuel consumption. This is where technologies

that determine optimal ship routes based on accurate

weather data come in. These technologies contribute to

reducing fuel consumption, thus reducing emissions

and costs, while enhancing maritime safety by

avoiding hazardous weather conditions.

Ship weather routing has gained attention in both

academic and industrial circles. It primarily aims to

improve voyage profitability and ensure the safety of

ships and crews. Weather routing optimization

identifies the best ship routes, balancing conflicting

goals such as reducing fuel use-which may lead to

unsafe speeds in rough weather-and ensuring safety,

which might increase travel time and emissions. It also

facilitates the optimization of ship operational

parameters to achieve a balance between speed, fuel

efficiency, dynamic loading, and the influence of

weather conditions. The goal is to minimize voyage

time, fuel usage, and safety risks, rather than to

increase speed directly (which is limited by wave

impact) [6]. Voyage optimization helps shipping

companies to lower costs and improve

competitiveness. Efficient operation across cost,

energy, and safety is essential. Many vessels now a

days use weather routing systems, which rely on

meteorological and oceanographic data, ship

characteristics, and route information. Depending on

operator goals, optimization may focus on energy

efficiency, travel time, safety, or a mix of these. The

need for quick, effective decisions is growing due to

fast-changing weather and navigation conditions. This

calls for advanced computational methods to improve

efficiency and support fast, reliable planning [7].

The main goals of weather routing are often in

conflict, making MOO crucial to find the best trade-

offs. This process is resource-intensive and requires

analyzing large datasets—ranging from weather

forecasts and vessel specifications to environmental

limits. Modern routing solutions aim to reduce travel

time and costs while improving sustainability and

resource use. These solutions depend on optimization

algorithms that can handle both environmental and

operational constraints, especially under dynamic and

unpredictable weather. A common method in the

literature is to model routing as an MOO problem [8],

incorporating both static (land and shallow waters)

and dynamic (e.g. regions of high waves) constraints.

The challenge increases with uncertainties in weather

and vessel behavior, which make the optimization

process far more complex. Metaheuristic algorithms

are widely used to solve complex, constrained [9], and

nonlinear MOO problems. Although effective, the

optimization of ship routes under weather constraints

remains computationally intensive, posing challenges

for real-time or near-real-time applications.

Recently, High-Performance Computing (HPC) has

become important for solving complex problems in

science, engineering, and data analysis. It enables faster

simulations, supports breakthroughs in Artificial

Intelligence (AI), and helps researchers process

massive datasets. The success of HPC is counting on

multidisciplinary efforts, including supercomputers,

parallel programming, network-based HPC

environments, and HPC applications. HPC uses

advanced systems with parallel processing (e.g.,

Compute Unified Device Architecture (CUDA), Open

Multi-Processing (OpenMP), Message Passing

Interface(MPI) to solve complex, data-heavy

problems much faster[10]. Originally used in scientific

research, HPC is now applied in many computer

science and engineering fields.

The ship weather routing models need to process

huge amounts of data and a heavy workload with the

requirement of ultra-fast response. Therefore, it is of

extreme necessity to promote computation efficiency

by applying HPC to ship weather routing systems.

Using HPC allows the design of scalable and efficient

frameworks to manage the growing complexity of

weather routing.

This paper tries to outline how HPC could be

integrated with multi-objective optimization for more

efficient ship routing. It aims to show how HPC can

improve the performance and practical use of routing

systems. The rest of this paper is structured as follows:

the next two sections discuss related work on Ship

Weather Routing (SWR), including its Multi-Objective

Optimization (MOO) variants, and High Performance

Computing (HPC). The following section introduces

the few existing s.o.t.a. research on combined SWR and

HPC. Then in the next section we identify research

gaps and the need for further research for utilizing

HPC in this context. The following section outlines a

framework for HPC-based SWR solution. Finally, the

last section concludes the material presented,

including also some recommendations for future

research.

2 SHIP WEATHER ROUTING (SWR)

Voyage planning is a fundamental part of safe ship

navigation. It helps avoid risks in areas with high

maritime traffic, shallow waters, or unfavorable

weather conditions[11]. The IMO requires a

comprehensive voyage plan approved before

departure. This plan must address routing systems,

environmental protection, and safety in line with the

Safety of Life at Sea (SOLAS) convention[12]. Routing

systems—such as Traffic Separation Schemes (TSS),

precautionary areas, and zones to avoid—support

navigational safety and environmental compliance.

However, static hazards alone do not define routing

challenges. Dynamic meteorological and

oceanographic conditions also play a major role, with

waves, winds, and currents directly affecting fuel use,

speed, and route safety.

Ship weather routing research typically falls into

two groups: global path planning, which seeks an

optimal route between origin and destination based on

environmental data, and local path planning, which

focuses on short-term adjustments. Early log-term

(global) weather routing methods relied on the

isochrone algorithm introduced in 1957 [13]. It

calculates the furthest distance a ship can travel in

equal time intervals, assuming fixed environmental

conditions. While simple and intuitive, when

implemented as a computer algorithm it introduced

issues such as isochronal loops. Hagiwara [14]

addressed this by using sector-based waypoint

expansion and pruning suboptimal subsectors. Later,

Lin [15] proposed a three-dimensional variant using a

dynamic grid, and Chen et al. [16] improved it further

with the Isochrone-A* method, which dynamically

adjusts the search space and uses an augmented cost

function for multi-objective optimization.

With the rise of digital charts and computational

capacity, graph-based algorithms such as Dijkstra’s

[17] have become widely adopted. These algorithms

model sea areas as graphs, with edge weights adjusted

for weather-induced speed losses. Sen and Padhy [17]

introduced weights based on reduced speeds due to

wind and waves. Takashima et al. [18] adapted

Dijkstra’s algorithm for fuel savings in coastal

navigation by adjusting propeller revolutions.

However, standard Dijkstra paths are often not smooth

and may lack flexibility. Modifications have

introduced features such as collision avoidance[19],

MOO [20], and time-dependent variables. These

enhancements improve applicability but do not resolve

limitations in path realism and continuous control.

Dynamic programming (DP) offered an alternative,

particularly for grid-based optimization. In early DP

approaches to weather routing De Wit[21] and Calvert

et al. [22] applied two-dimensional DP (2DDP) to

optimize route selection under static ship settings. This

approach divides the sailing region into a spatial grid

and uses recursive optimization over time steps. To

capture more realistic behaviors, later studies extended

this to three-dimensional DP (3DDP), including control

variables such as speed and engine power. Wei and

Zhou[5] reduced the high computational demand of

3DDP by clustering similar control states. Still, DP-

based methods remain demanding, especially when re-

optimization is needed due to real-time changes in

weather or vessel status.

3 MULTI-OBJECTIVE OPTIMIZATION (MOO) IN

THE CONTEXT OF SWR

In recent years, Multi-Objective Optimization (MOO)

has gained more attention, also in the context of ship

weather routing, due to its ability of balancing the

complex trade-offs between conflicting goals (in SWR:

fuel efficiency, voyage time, emissions, and safety).

While traditional methods remain as foundational

approaches, they have been adapted to handle

multiple goals- modified classical algorithms that

generate Pareto-optimal solutions to advanced

evolutionary algorithms tailored for simultaneous

optimization of time, cost, safety, and emissions. In this

context, MOO—particularly in hybrid forms—has

demonstrated effectiveness, the gains associated with

the speed of classical methods with the flexibility of

modern algorithms. These hybrid models usually rely

on traditional methods such as Dijkstra[23], isochrones

or dynamic programming to generate initial routes or

perform local searches while using metaheuristics to

explore complex solution spaces.

Numerous MOO-based approaches and hybrid

strategies have been introduced to determine the

optimal route and support solution diversity. For

example, improved multi-objective ant colony

optimization (IMACO) [24] combined with the

Technique for Order Preference by Similarity to Ideal

Solution (TOPSIS) has shown potential in balancing

navigation risk and fuel consumption. Similarly,

hybrid versions of Particle Swarm Optimization (PSO)

and Genetic Algorithms (GA), enhanced with

heuristics, have been used to optimize ship fuel

consumption mode and routes under environmental

constraints[25]. Researchers have also customized the

Estimation of Distribution Algorithm (EDA) [26] for

planning ship routes in a non-stationary

environment (tidal water) with multiple goals by using

probabilistic models of the best solutions instead of the

usual crossover and mutation methods. In practice,

EDA explores the solution space by sampling, selecting

strong route candidates, and building probability

distributions that reflect their key traits. This method

avoids early convergence to weak solutions by keeping

a diverse set of options, handles trade-offs between fuel

and time more effectively, and adapts well to uncertain

weather. Evolutionary algorithms, especially the Non-

dominated Sorting Genetic Algorithm II (NSGA-

II)[27], have been also applied. These methods provide

Pareto-optimal solutions that help operators select

routes based on specific preferences. The Multi-

Objective Evolutionary Algorithm based on

Decomposition (MOEA/D) further improves

performance by dividing the problem using weight

vectors. An advanced version, w-MOEA/D[28] which

applies w-dominance[29] to standard MOEA/D

algorithm, uses ensemble forecasts for uncertainty

management, and accounts for real-world constraints

such as bathymetry and emission zones. Despite these

advantages, applying an MOO-based and hybrid

approach in ship weather routing is computationally

high and time-consuming. Evaluating multiple

objectives requires searching a large multidimensional

solution space and running various scenarios under

varying weather conditions, which can significantly

increase computation time. As a result, real-time or

near-real-time application of MOO remains difficult

without additional support.

Recent studies have introduced also Machine

Learning (ML) and hybrid methods to improve

weather routing performance. ML has been used to

enhance heuristic and meta-heuristic functions in

classic algorithms, guiding the search toward better

candidate routes. For example, Chen and Mao [30]

incorporated ML-based cost predictions into the

Isochrone method to improve waypoint selection and

route smoothness. Moreover, Shin et al.[31] proposed

an improved A* algorithm for ship routing that

integrates AIS and marine weather data to address

limitations in existing methods. They introduced an

adaptive grid system to reduce computational

complexity and applied a 16-way search strategy to

generate smoother and more economical routes.

Machine learning models were used to estimate the

ETA and evaluate route efficiency. Simulation results

showed that their proposed method outperformed

traditional A* routing by providing more economical

paths. Hybrid methods often combine exact or

heuristic algorithms with metaheuristics to balance

quality and speed. In this regards, Qian et al. [32]

integrated A* with genetic algorithms to handle speed

variations over a dynamic grid. Similarly, Ma et al. [33]

combined A* with an improved NSGA-III to support

MOO under environmental and operational

constraints. These approaches leverage the strengths of

different algorithms—using graph search for reliable

structure and evolutionary methods for flexible

optimization. However, their performance still

depends on the quality of initial solutions and the

tuning of ML models, especially under uncertain and

changing sea states.

Despite these improvements, no single method

provides an optimal solution covering all cases.

Heuristic and graph-based approaches provide speed

but often ignore multi-objective trade-offs. Dynamic

Programming (DP) provides global optima; however,

it sacrifices scalability. Metaheuristics handle multiple

objectives and constraints but can suffer from long

convergence times and uncertain performance in

dynamic settings. To benchmark and improve these

approximations, VISIR [34] and its successor[35]

provide reference solutions by solving the governing

equations of motion under realistic weather conditions.

VISIR-2 improves the model by adding factors such as

wave angles, wind and current directions, and fuel

emissions, which helps in finding routes that produce

the least CO₂ and assists researchers in checking how

accurate the approximate methods are. Recently,

researchers' efforts have shown that ship weather

routing has progressed beyond simple time or fuel

minimization to advanced multi-objective

optimization. Modern methods integrate route

planning with control parameters such as speed,

power, and trim while accounting for changing

weather conditions and vessel behavior. Higher-

resolution environmental data and ensemble forecasts

have improved accuracy but increased the size of input

data[28]. As a result, algorithms must handle larger

solution spaces and evaluate more variables, which

leads to higher memory usage and longer runtimes.

This requirement is especially critical in methods such

as 3D dynamic programming or population-based

algorithms that evaluate many control states at each

step. Machine learning and hybrid approaches arise to

guide search and improve performance. However,

implementation of it in real-time use remains limited

due to computational costs and the need for dynamic

updates during long voyages. Clean fuel ships and

new propulsion types have more variables to optimize,

which makes the problem even harder. Benchmark

studies are still rare, which makes comparing methods

and validating improvements difficult.

A potential solution able to solve the complexity,

high dimensions, and memory requirements of MOO

problems in ship weather routing is HPC. It provides

an approach to splitting tasks across many processors

or GPUs; HPC reduces run times and manages the

increasing memory requirements. It lets models

evaluate more route options and weather scenarios in

near real time.

4 OVERVIEW OF HIGH PERFORMANCE

COMPUTING (HPC)

Recently, HPC has become an essential tool across

scientific communities. A vast increase in the use of

HPC has been observed since its origins in the mid-

20th century, and it has become essential in modern

systems. Researchers struggle with computational

problems when their problem needs a huge

computation and memory resources. Since typical

desktop computers do not work efficiently compared

with HPC, the result is usually low-performance

computing. HPC plays a key role in accelerating

computational tasks and enabling large-scale data

analysis across various fields. In practice, the concept

of accelerating computation includes load balancing

across processors, optimized storage systems, and

efficient data processing. In this regard, the term

speedup in this context (e.g., in parallel computing),

refers to the speed at which a task can operate when

divided among multiple processors. Amdahl’s Law

clarifies that the non-parallel part of a program sets a

limit on the total gain. For example, if 10% of the

process cannot truly run in parallel (the process runs

only sequentially), no amount of added processors can

reduce the overall time below that 10%. This

phenomenon makes sequential computing processes—

tasks one after the other—a bottleneck. A bottleneck is

something that slows down the overall process—such

as a narrow point in a bottle that restricts flow.

Moreover, the serial code represents the narrow point.

In contrast, Gustafson’s Law justifies that scaling the

problem size with the number of processors can keep

them fully utilized. As a result, a complex problem can

be solved in the same time by using more processors,

which helps reduce the impact of the serial portion.

These ideas relate to strong and weak scaling in

HPC; the terms "weak scaling" and "strong scaling"

refer to how well a system can maintain its

performance as the number of processors or nodes is

increased. Strong scaling looks at performance when

the problem size stays fixed but the processor count

grows. Weak scaling tests if performance holds steady

when both the problem size and processor count grow

together. These model help guide parallel algorithm

design by highlighting trade-offs between speed,

efficiency, and processor usage. That explains why

HPC is considered a major leap in computing, offering

the power to solve complex problems more efficiently

and unlock new capabilities. HPC architecture takes

many forms based on users' needs. Researchers can

choose different ways to design HPC systems.

Designing an HPC system may involve a combination

of parallel computing, cluster computing, and

grid/distributed computing strategies. HPC

architectures are commonly categorized into three

main components: compute, network, and storage.

These components work together to enable high-speed

processing of complex tasks.

Beyond these, systems also differ in how processors

and memory interact. Memory-architecture models fall

into three categories:

1. Shared-memory systems

All processor cores access a single, global address

space. Threads coordinate through lock or barrier

primitives and share data via common variables.

2. Distributed-memory systems

Each computing node has its own private memory.

Nodes exchange data by passing explicit messages

over an interconnect (for example, using MPI).

3. Hybrid systems

Nodes combine multicore CPUs and accelerators

(GPUs) and use both shared-memory and message-

passing techniques. Within a node, threads share

memory; between nodes, processes communicate

by messaging.

Each model aligns with different programming

frameworks (e.g., OpenMP for shared memory, MPI

for distributed memory, CUDA or OpenCL for GPU

acceleration). Efficient use of shared, distributed, and

hybrid memory architectures depends on choosing the

right parallel framework. The following table (Table 1)

summarizes common frameworks and their typical use

cases.

Table 1- HPC Programming Frameworks

Framework

Model

Target

Hardware

Language

Support

Use Case

MPI

Distributed

CPU clusters

C, Fortran,

Python

Large-scale

messaging

OpenMP

Shared

Multi-core

CPU

C, C++,

Fortran

Loop-level

parallelism

CUDA

GPU

NVIDIA

GPUs

C/C++

Data-parallel

tasks

OpenCL

Heterogeneous

GPUs, CPUs

Portable

kernels

OpenACC

Directive-

based

GPUs

C, Fortran

Simplified

GPU use

As mentioned above, the HPC relies on splitting

tasks into smaller parts that can run at the same time

through applying several technologies that are used to

implement and create high-performance computing

systems. Parallel processing is a method of running

more than one processor to handle separate parts of a

single task. Dividing different parts of a task among

multiple processors minimizes the time required,

achieves lower power consumption, and enables more

efficient multitasking of the program.

Most systems with more than one central

processing unit (CPU) are able to perform parallel

processing, in addition to the multi-core processors

commonly found on computers today. Another

method uses GPUs, which contain thousands of

lightweight cores designed to handle many repeated

tasks in parallel. Moreover, SIMD (Single Instruction,

Multiple Data) is another technique used in CPUs; it

allows one instruction to process many data points at

once, often speeding up loops. Some advanced models

have incredible portability across various computer

architectures and operating systems, such as MPI

(Message Passing Interface). MPI allows several

computers to work simultaneously on a problem by

exchanging data across a network. MPI is common in

large simulations run on clusters. On the other hand,

the OpenMP API supports multi-platform shared-

memory parallel programming in C/C++ and Fortran.

The OpenMP API defines a portable, highly scalable

model.

Furthermore, heterogeneous systems combine

CPUs, GPUs, and other accelerators. This setup

balances flexibility and speed by allowing each device

to handle the parts it performs best on. These models

form the core of HPC and make it possible to solve

complex problems faster and at larger scales.

OpenMP and PThreads are used for multithreading

on shared memory architectures. CUDA and OpenCL

enable GPU programming, whereas OpenACC

simplifies the use of GPUs through compiler directives.

MPI is the de facto standard for message passing,

whereas OpenSHMEM provides a shared memory

programming model in distributed systems. Libraries

such as Apache Hadoop and Spark support large-scale

data processing based on a task-based and dataflow

abstraction model. HPC programming requires trade-

offs between low-level control and high-level

abstraction. Lower-level interfaces such as MPI offer

fine-grained control at the expense of requiring more

intricate development. Higher-level programming

models simplify development, possibly at the cost of

performance. Recent developments indicate a shift

toward hybrid parallelism, energy-efficient

computing, and emerging memory technologies such

as NVRAM. These trends aim to increase efficiency and

better suit emerging workloads such as AI applications

and real-time analytics.

4.1 High-performance computing (HPC) architectures

HPC architectures include a broad range of hardware

configurations designed to solve complex problems

and perform intensive computational tasks.

Traditional HPC clusters usually consist of

homogeneous compute nodes connected through

high-speed, low-latency. HPC systems support

multiple layers of parallelism, from instruction-level

parallelism within processor cores to node-level

parallelism across clusters. One can then organize them

as a hierarchical structure of parallelism categories that

spans from the lowest level (instruction-level within a

single core) through intermediate levels (thread-level,

core-level) to the highest levels (node-level, cluster-

level). Along with that, most HPC systems also contain

complex memory hierarchies, including several cache

levels, local memory, and often shared or distributed

memory. Shared memory within a node enables rapid

data exchange between threads but requires careful

coordination to avoid race conditions. This architecture

demands close attention to data locality and movement

since performance degrades quickly when data must

travel farther from the processor. Figure 2 presents a

three-tiered view of HPC architectures that includes, at

the top, the Parallel Programming Model layer, which

encompasses message-passing (MPI) for process-level

coordination and shared-memory/threading

approaches (OpenMP/PGAS) for intra-node execution;

the middle Middleware layer comprises

communication libraries (MPI/PGAS/RDMA), job

scheduling and resource managers (e.g., Slurm,

Torque, Sun Grid Engine), and parallel file systems

(Lustre, NFS, GPFS) that connect compute tasks and

data movement across nodes; and at the base, the

Infrastructure layer includes physical compute engines

(CPUs, GPGPUs, FPGAs), network fabrics

(Infiniband/Ethernet, iSCSI/FC), operating systems

with numerical and accelerator libraries

(BLAS/LAPACK, CUDA/OpenCL), and centralized

storage arrays (SAN/NAS), illustrating how hardware,

system software, and programming frameworks

interlock to drive large-scale scientific applications.

Recently, graphical processing units (GPUs) have

become a major component of modern HPC systems.

The emergence of Graphics Processing Units (GPUs)

transformed HPC architecture. Originally built for

rendering graphics, GPUs evolved into general-

purpose accelerators. With thousands of simple cores,

they handle data-parallel workloads efficiently. When

aligned with suitable algorithms, GPUs can greatly

outperform CPUs. Today, many of HPC computer

systems are characterized by their both CPUs and

GPUs to match different computational demands. This

hybrid model boosts performance in fields such as

deep learning, molecular dynamics, and

computational fluid dynamics and computational fluid

dynamics. However, software must be tuned to the

GPU’s execution model to exploit its full potential.

Figure 2. HPC architecture[36]

Widely used frameworks such as MPI, OpenMP,

and CUDA provide fine-grained control over

resources. Moreover, alternatives techniques such as

OpenCL and Heterogeneous-compute Interface for

Portability (HIP) enable code portability across

different hardware. Other techniques, including Data

Parallel C++ (DPC++) and C++ Single-source

Heterogeneous Programming for OpenCL (SYCL),

offer modern C++-based models for writing portable

parallel code. Directive-based approaches such as

Open Accelerators (OpenACC) simplify GPU adoption

without major code changes. Additional libraries such

as Performance Portability Programming EcoSystem

(Kokkos) and RAJA Performance Portability Layer

(RAJA) support writing portable code across CPUs and

GPUs. Such programming models help developers

optimize hybrid HPC systems while maintaining

flexibility across architectures.

4.2 Implementation of High-Performance Computing in

different domain

Today, HPC systems are revolutionizing scientific

research by providing the power needed to tackle

complex problems across many fields. These systems

let researchers process large datasets, run detailed

simulations, and build predictive models with

unmatched speed and accuracy. Engineers use it for

mechanical design, device testing, and product

development, minimizing costs and increasing

efficiency. In the automotive sector, HPC drives digital

simulations and safety assessments. Energy companies

apply it to reservoir modelling and subsurface geology,

evaluating resource potential and estimating costs.

Despite its broad adoption, HPC remains

underused in maritime applications—especially ship

weather routing. Most routing systems still rely on

sequential programming or on low-parallelism

methods that have been implemented on some

algorithms, such as genetic algorithms or graph-based

searches (A*, Dijkstra). On the other hand, related

maritime domains, such as coastal flooding, vessel

tracking, and ship-ice modelling, have seen some HPC

use, but ship routing lags. Ship routing under dynamic

weather conditions demands frequent recalculations of

forecasts, wave states, wind directions, and ship

constraints.

Recent studies highlight HPC’s potential impact on

maritime research. For example, a researcher in

cooperation with the National Supercomputing Centre

ran full-order fluid–structure interaction simulations

on a 14.4 million-cell [37]. They used the PETSc library

— an open-source toolkit for solving large-scale

scientific problems— and a GMRES solver to solve big,

complex problems at the same time on five computers,

(typical multi-core desktop ones, with 24 cores, using

Intel® Xeon® E5-2690 v3 CPUs (2.60 GHz). This

configuration reduced runtimes from days to mere

hours. It let researchers sweep through dozens of

design parameters in one campaign—something

impossible on smaller clusters. By handling high cell

counts and fine time steps, the simulations captured

millimeter-scale wave features and non-linear wave–

wave interactions. Building on these simulations, the

team developed data-driven digital twins of ships and

offshore structures. The twins run in near real time,

ingesting live environmental data to predict structural

responses under changing loads. Such examples

emphasize the ability of HPCs to run ensemble

forecasts, whereas uncertainty quantification lets

operators explore multiple scenarios in minutes. This

capability supports the adaptive control of wave

paddles in test basins and, ultimately, onboard

decision support for vessels at sea. Yet most ship

routing systems stick to legacy solvers, even though

modern parallel methods could boost performance;

they still rarely use these advances. The enhanced ship

routing model should implement HPC strategies such

as combining distributed computing for broad tasks,

shared-memory threads for local work, and hardware

accelerators for compute-intensive operations.

The maritime sector now faces rapid changes due to

digitization, decarbonization, and safety demands. As

ships tap real-time forecasts and IoT data, computing

needs will surge. HPC could address these needs by

enabling live, multi-objective route planning that

balances safety, fuel use, and transit time. Traditional

methods remain dominant in current systems;

enhanced parallel computing approaches offer

significant untapped potential. However, it should be

noted that developing HPC solutions is not easy; it

demands deep knowledge of parallel programming,

memory management, and hardware architecture. To

fully utilize HPC, developers must create routing

algorithms that are specifically designed to take

advantage of the latest parallel hardware and

programming models. Many developers avoid HPC

methods because they demand deep expertise in

parallel programming and system optimization.

Writing code for multi-level parallelism requires

understanding of memory models, concurrency, and

hardware architecture. These challenges raise

development costs and extend project timelines as a

result, teams stick to simpler sequential or limited-

implementation.

5 SHIP WEATHER ROUTING TECHNIQUES

BASED ON HPC

Multi-objective ship weather routing balances safety,

fuel use, and voyage time by treating each as an

optimization objective. The process evolves route

candidates over many generations. Each route is

evaluated on passage time, fuel consumption, and

safety, yielding a set of non-dominated solutions

known as the Pareto front. The voyage is divided into

stages, and at each stage the algorithm records the best

trade-off among objectives for each state. This process

builds a map of optimal choices. To compare options,

the objectives might be combined into a single score

using user-defined weights, but thus the solution may

overlook some Pareto-optimal routes.

Ship weather routing is computationally complex

because environmental conditions change dynamically

(in space and time). Wind, waves, and currents might

be represented on a high-resolution grid. Each grid

point holds forecasts for certain number of hours.

Evaluating a route requires sampling forecasts at each

step and multiplying data readings. As a result,

datasets can span thousands of latitude–longitude cells

over dozens of time steps and require gigabytes of

RAM. Tracking constitutes hundreds of route

candidates across multiple weather scenarios can push

memory needs into tens of gigabytes. These demands

make the problem a clear fit for HPC. Indeed, HPC

meets these demands by rapidly processing large

datasets and executing routing algorithms in parallel.

It supports the conflicting goals of travel time, fuel use,

and safety under unpredictable sea conditions [38].

Using HPC, one can evaluate numerous route options

with high-resolution environmental data in real time

[39]. Traditional serial methods cannot meet this

requirement. A key advantage—especially with GPU

acceleration—is parallelizing route evaluations. For

example, implementing Non-local Kernel Density

Estimation (NLKDE) on GPUs enabled real-time

trajectory visualization of AIS data [40]. Similar

methods can be applied to ship routing by evaluating

cost functions over thousands of paths using

multithreading, GPU or FPGA accelerators, and SIMD

architectures. Models such as OpenMP and CUDA

support maritime applications, allowing tasks such as

segment-based risk analysis or energy prediction to

run in parallel.

Integrating HPC into ship routing reduces

simulation time, improves accuracy through higher-

resolution data, and supports real-time evaluation of

diverse options and constraints. It also addresses

growing computational needs in the maritime sector. A

notable example is the Sailing Assistance Application

(SAA) by Życzkowski et al[19]. the first parallel

implementation of deterministic weather routing for

sailboats. Deployed in a cloud environment with

OpenMP, SAA achieved up to 79.1% speedup in

routing components while preserving accuracy. It

divided tasks into navigation area generation and

sailing condition modeling using shared-memory

architectures. However, SAA focuses on sail-assisted

vessels, not the dominant mode of commercial

shipping – and uses Dijkstra’s algorithm, which is less

optimal for large-scale or multi-objective problems.

Their proposed method implements HPC techniques

only through OpenMP, which means it does not take

advantage of other HPC techniques. Choosing HPC

techniques requires analyzing various options, such as

GPU acceleration or distributed memory models

(MPI), which makes the proposed approach more

difficult to scale for large areas and high-resolution

grids.

To generalize and improve HPC-based routing, one

might adopt a vessel model that supports a generalized

engine-driven ships. A multi-level parallelism

framework combining MPI, OpenMP, and GPU

acceleration would enhance scalability.

100

Algorithmically, parallelized multi-objective solver

can better match real-world needs. Dynamic rerouting

and adaptive grid resolution would enable responsive

and efficient navigation. These enhancements would

yield a versatile, high-performance system across

vessel types and scenarios. As presented in another

study[32] Segment Parallel A* (SPA*) was presented.

SPA* applies HPC principles to optimize maritime

path planning under complex environmental

conditions. It divides the route into sub-routes and

distributes computation across multiple CPU cores

using MATLAB.

Each segment uses an enhanced A* algorithm that

balances risk, distance, and speedups from (as reported

by the Authors) five to over 12,425 times compared to

Dijkstra, A*, Bidirectional A*, Ant Colony

Optimization, Harris Hawks Optimization, and

Sparrow Search Algorithm. SPA* efficiently handles

high-resolution, multi-constraint data, including real-

time wind, wave, and topographic risk inputs. Yet it

relies solely on CPU parallelism and grid-based

models, which may incur overhead at high resolutions

or from excessive segmentation.

Most prior ship routing methods remain sequential,

even though parallel techniques have been applied in

related domains such as swarm-based optimization

[41][42][43][44], graph computations [45], and robotic

motion planning [46][47]. Integrating HPC with

deterministic routing may reduce execution time, but it

limits flexibility for multi-objective and real-time

applications. In contrast, parallelizing metaheuristic

MOO algorithms—such as NSGA-II, NSGA-III, or

MOEA/D—offers a scalable alternative for full-scale

ship weather routing. These methods can better exploit

HPC, support dynamic constraints, and enable faster

decision-making with broader optimization goals. A

deterministic approach, in this context, has limited

applicability.

6 THE LITERATURES GAP

The literature review presented in the previous

sections revealed that ship weather routing has gained

renewed interest recently as regulators and operators

press to cut emissions and boost efficiency. Scholars

have shifted their focus to ship weather routing, which

employs multi-objective optimization to balance

competing goals under varying weather conditions.

But multi-objective ship weather routing significantly

increases computational complexity compared to

single-objective optimization. The multi-dimensional

optimization space generates exponential growth in

computational requirements; it often exceeds the

available decision windows in practical navigation

scenarios. This "curse of dimensionality" becomes

particularly problematic when considering high-

resolution environmental data needed for accurate

routing. Moreover, ship weather routing requires

timely decision-making, especially in dynamic weather

conditions, which create significant limitations for

providing real-time solutions to complex multi-

objective routing problems. State-of-the-art methods

can find globally optimal solutions under specific

conditions. However, recent reviews highlight the

increasing complexity of these methods as it handle

multiple objectives, which increases computational

demands[48]. While they offer strong guarantees, their

efficiency drops with large-scale or dynamic problems.

This becomes a key limitation in ship routing, where

weather and ocean conditions change frequently and

require continuous re-optimization.

These challenges have revealed HPC’s potential for

overcoming them and paved the way for broader use.

Despite these developments, HPC remains

underutilized in weather-based ship navigation so far.

The literature review identified several significant

research gaps at the intersection of HPC and multi-

objective ship weather routing. Many ship routing

algorithms do not effectively utilize the HPC

techniques, such as parallel computing capabilities,

multi-core and other HPC techniques and resources.

Moreover, there is a lack of standardized frameworks

(scalability tests, varying core counts, and problem

sizes to show how runtime changes) to evaluate and

compare HPC performance in ship routing

applications for weather, hindering systematic

improvement in the field. Limited discussion exists on

how HPC resources should scale to accommodate

increasing complexity in multi-objective optimization

scenarios. The literature inadequately addresses how

existing ship weather routing can overcome time

constraints in dynamic weather routing scenarios

requiring near real-time decision-making.

Additionally, many existing methods for ship weather

routing, such as pathfinding algorithms, depend on a

step-by-step process that creates links between each

step, making it difficult to run them at the same time.

Thus, there is a clear need to assess ship weather-

routing algorithms for implementation on high-

performance computing platforms. This assessment

must identify which steps can run in parallel, such as

route evaluation, weather-data interpolation, and ship-

performance prediction. The existing ship weather

routing models face scalability limitations when

handling global-scale routing problems with high-

resolution environmental data and multiple

optimization objectives. Furthermore, challenges in

managing HPC resources in shared environments are

evident but not thoroughly addressed in the literature.

The computational challenges of processing large

volumes of dynamic environmental data in HPC

environments remain under-explored, and current

ship-routing algorithms exhibit limited parallelization.

The inherent sequential nature of many route

planning algorithms rely on sequential pathfinding

routines that enforce dependencies between steps and

hinder concurrent execution. Although some processes

could run in parallel, some of the core ship weather

routing algorithmic structure demands sequential

processing, weakening the benefits of multi-core

systems.

Recently, Machine Learning (ML) empowered ship

weather routing has shown promise due to the

emergence of ML and its role in various domains;

however, there is insufficient research on effectively

integrating these approaches with HPC for ship

routing applications. ML models potentially address

some of the inherent limitations of traditional

optimization algorithms, but the computational

demands of training and using complex models create

additional challenges for HPC integration.

101

To address the existing challenges, we aim to

outline a proposal based on implementation of HPC in

the existing MOO-SWR algorithm.

7 THE PROPOSED APPROACH TOWARDS SHIP

WEATHER ROUTING WITH MOO AND HPC

Despite major improvements, current methods still do

not have a single system that can manage both engine-

driven vessels and detailed grid models, particularly

when the algorithm’s execution time is limited. To

address the existing challenges, we aim to propose a

solution based on the implementation of HPC in the

existing MOO-SWR algorithm.

The proposed approach begins with a review of

recent literature and practical case studies to identify

algorithms with strong potential for HPC

implementations. Simultaneously, we’ll analyze

existing MOO-SWR methods to pinpoint components

suitable for HPC techniques—such as parallelization

across processors, shared-memory architectures, or

specialized hardware—and then design and execute

benchmarks to measure CPU/GPU utilization, memory

usage, solution quality, and speedup relative to

sequential baselines on identical hardware/software

stacks.

Next, we’ll evaluate these algorithms across various

HPC software frameworks and hardware

configurations, comparing their performance to

sequential implementations. For example, to maximize

efficiency, we may adopt a hybrid CPU–GPU

architecture: GPUs will handle parallel tasks (e.g.,

environmental data interpolation and route-cost

evaluation), while CPUs manage sequential route-

optimization steps, thereby accelerating multi-

objective optimization and containing computational

growth.

The proposed solution will also explore integrating

an ML-based forecasting model—trained on historical

and real-time data—to predict short-term sea and

weather conditions. These predictions will trigger

targeted re-optimizations within the routing

algorithm, avoiding full re-computations when

changes are localized. Together, these measures

enhance speed, accuracy, and resilience in ship

weather routing.

Figure 3 presents part of the proposed solution,

which ensures the ability to provide reliable route

recommendations promptly. According to Figure 3, the

framework gathers dynamic weather data, ship

parameters, and business information. It maps the

ship’s speed and course according to time, cost, and

CO₂ objectives. The framework then selects an MOO-

SWR algorithm based on its capability to operate on

HPC systems. Using the selected algorithm, it chooses

processes that support HPC features—such as parallel

processing—and specifies the hardware and software

methods for the chosen solver. These methods may

include genetic algorithms or distributed gradient

searches, chosen at runtime based on solution quality.

Finally, it compares solver performance, route

efficiency, and resource use to determine the optimal

HPC approach, routes, and metrics.

Figure 3. MOO-SWR-based HPC

8 CONCLUSIONS

Ship weather routing has emerged as a critical area of

research and application in maritime transportation,

with significant potential to improve safety, reduce

fuel consumption, and minimize environmental

impacts. The multi-objective-based HPC approach

builds on the importance of weather-aware voyage

planning and provides the computational muscle and

methodological sophistication needed to turn routing

into practice in real-world scenarios.

Accordingly, using High Performance Computing

(HPC) for Multi-Objective Optimization (MOO) in

weather-based ship routing is a major advance over

simple, deterministic methods. Deterministic methods

are fast and easy to use. In contrast, multi-objective

methods balance multiple goals in complex, changing

sea conditions. The pure MOO methods face

challenges such as a multidimensional solution space,

trade-off analysis, detailed environmental models,

safety integration, high compute needs, data fusion

and management, real-time updates, and strict

verification and validation. Conversely, the HPC-

supported MOO methods promise to solve the

complex optimization problems and open

opportunities for addressing these challenges with cost

savings and more flexibility.

HPC may use different strategies and resources to

achieve significantly more computing power than a

standard computer. It relies on a combination of

specialized hardware and software to enable complex,

data-intensive applications to run at high speeds and

allow for increased efficiency. HPC involves

aggregating resources, such as supercomputers,

clusters of computers, and others, to enable the parallel

processing of complex calculations across numerous

processors. Moreover, it offers a suite of strategies that

utilize computational power for various tasks. These

strategies can involve expanding computing resources;

for instance by using parallel processing with MPI and

OpenMP to distribute calculations across many cores.

GPU acceleration with CUDA enables dealing with

ensemble simulations for diverse weather and routing

scenarios. Task-based frameworks and hybrid CPU–

GPU architectures handle real-time adaptation and

data fusion. These approaches help explore trade-offs,

validate solutions, and meet the demands of complex

routing. However, there is no guarantee from a

theoretical perspective that selecting the HPC strategy

from various options—such as MPI-based parallelism,

102

GPU acceleration, cloud-native clusters, or hybrid

models—will yield the best result. Thus, the selection

of suitable techniques requires analyzing and

comparing these approaches. That analysis demands

profound domain knowledge and strong

programming skills. Despite these hurdles, multi-

objective HPC methods can minimize fuel use, boost

safety, and lower the environmental impact. With

greater HPC availability and increasing computational

demands for algorithms, these methods might gain

broader use in weather-based ship routing. The MOO-

SWR drive for efficiency, safety, and sustainability will

spur further innovation in weather-based routing.

Addressing these challenges and exploring diverse

HPC strategies, researchers can build efficient models

based on HPC to resolve difficult trade-offs in high-

demand computational domains, such as maritime

domains, in general and particularly for ship weather

routing.

Future work will include searching for a validated

unified HPC-based multi-objective ship routing

framework that better matches real-world scenarios,

enabling quick responses under actual conditions. The

primary objective is to develop a highly efficient and

adaptable ship weather routing method capable of

handling large-scale environmental datasets and

rapidly updating route plans in response to changing

conditions, thereby ensuring reliable performance in

time-critical, real-time scenarios.

REFERENCES

[1] S. Deng and Z. Mi, “A review on carbon emissions of

global shipping,” Mar. Dev., vol. 1, no. 1, pp. 1–10, 2023,

doi: 10.1007/s44312-023-00001-2.

[2] ShipUniverse, “Top 30 Reasons Why Maritime Shipping

Is the Backbone of Global Trade,” 2024.

[3] J. Verschuur, E. E. Koks, and J. W. Hall, “Ports’ criticality

in international trade and global supply-chains.”

[4] Statista, “Breakdown of CO$_2$ emissions in the

transportation sector worldwide 2020, by subsector,”

2021. [Online]. Available:

https://www.statista.com/statistics/1185535/transport-

carbon-dioxide-emissions-breakdown/

[5] W. Shao, P. Zhou, and S. K. Thong, “Development of a

novel forward dynamic programming method for

weather routing,” J. Mar. Sci. Technol., vol. 17, no. 2, pp.

239–251, Jun. 2012, doi: 10.1007/s00773-011-0152-z.

[6] L. P. Perera and C. G. Soares, “Weather routing and safe

ship handling in the future of shipping,” Ocean Eng., vol.

130, pp. 684–695, 2017, doi:

10.1016/j.oceaneng.2016.09.007.

[7] L. Walther, A. Rizvanolli, M. Wendebourg, and C. Jahn,

“Modeling and Optimization Algorithms in Ship

Weather Routing,” Int. J. e-Navigation Marit. Econ., vol.

4, pp. 31–45, Jun. 2016, doi: 10.1016/j.enavi.2016.06.004.

[8] J. Szlapczynska and R. Szlapczynski, “Preference-based

evolutionary multi-objective optimization in ship

weather routing,” Appl. Soft Comput. J., vol. 84, Nov.

2019, doi: 10.1016/j.asoc.2019.105742.

[9] E. G. Talbi, M. Basseur, A. J. Nebro, and E. Alba, “Multi-

objective optimization using metaheuristics: Non-

standard algorithms,” Int. Trans. Oper. Res., vol. 19, no.

1–2, pp. 283–305, Jan. 2012, doi: 10.1111/j.1475-

3995.2011.00808.x.

[10] P. Czarnul, J. Proficz, and K. Drypczewski, “Survey of

Methodologies, Approaches, and Challenges in Parallel

Programming Using High-Performance Computing

Systems,” 2020, Hindawi Limited. doi:

10.1155/2020/4176794.

[11] Ü. Öztürk, M. Akdağ, and T. Ayabakan, “A review of

path planning algorithms in maritime autonomous

surface ships: Navigation safety perspective,” May 01,

2022, Elsevier Ltd. doi: 10.1016/j.oceaneng.2022.111010.

[12] K. SKÓRA and A. WOLSKI, “VOYAGE PLANNING,”

Sci. J. Silesian Univ. Technol. Ser. Transp., vol. 92, pp.

123–128, Sep. 2016, doi: 10.20858/sjsutst.2016.92.12.

[13] James and Richard W, Application of wave forecasts to

marine navigation. New York University, 1957.

[14] Hagiwara and Hideki, “Weather routing of (sail-

assisted) motor vessels.,” Technische Universiteit Delft,

1989.

[15] Y. H. Lin, M. C. Fang, and R. W. Yeung, “The

optimization of ship weather-routing algorithm based on

the composite influence of multi-dynamic elements,”

Appl. Ocean Res., vol. 43, pp. 184–194, 2013, doi:

10.1016/j.apor.2013.07.010.

[16] Y. Chen, W. Tian, and W. Mao, “Strategies to improve

the isochrone algorithm for ship voyage optimisation,”

Ships Offshore Struct., 2024, doi:

10.1080/17445302.2024.2329011.

[17] D. Sen and C. P. Padhy, “An approach for development

of a ship routing algorithm for application in the North

Indian Ocean region,” Appl. Ocean Res., vol. 50, pp. 173–

191, 2015, doi: 10.1016/j.apor.2015.01.019.

[18] K. Takashima, B. Mezaoui, and R. Shoji, “On the fuel

saving operation for coastal merchant ships using

weather routing,” Mar. Navig. Saf. Sea Transp., vol. 3, no.

4, pp. 431–436, 2009, doi: 10.1201/9780203869345.ch75.

[19] M. Zyczkowski and R. Szlapczynski, “Collision risk-

informed weather routing for sailboats,” Reliab. Eng.

Syst. Saf., vol. 232, no. November 2022, p. 109015, 2023,

doi: 10.1016/j.ress.2022.109015.

[20] C. U. Bottner, “Weather routing for ships in degraded

condition,” in Proceedings of the International

Symposium on Maritime Safety, Security and

Environmental Protection. Athens, Greece, 2007.

[21] S. Calvert, E. Deakins, and R. Motte, “A Dynamic System

for Fuel Optimization Trans-Ocean,” J. Navig., vol. 44, no.

2, pp. 233–265, 1991, doi: 10.1017/S0373463300009978.

[22] C. de Wit, “Proposal for Low Cost Ocean Weather

Routeing,” J. Navig., vol. 43, no. 3, pp. 428–439, 1990, doi:

10.1017/S0373463300014053.

[23] H. Wang, W. Mao, and L. Eriksson, “A Three-

Dimensional Dijkstra’s algorithm for multi-objective ship

voyage optimization,” Ocean Eng., vol. 186, no. June, p.

106131, 2019, doi: 10.1016/j.oceaneng.2019.106131.

[24] J. Yang, L. Wu, and J. Zheng, “Multi-Objective Weather

Routing Algorithm for Ships: The Perspective of Shipping

Company’s Navigation Strategy,” J. Mar. Sci. Eng., vol.

10, no. 9, 2022, doi: 10.3390/jmse10091212.

[25] W. Zhao, H. Wang, J. Geng, W. Hu, Z. Zhang, and G.

Zhang, “Multi-Objective Weather Routing Algorithm for

Ships Based on Hybrid Particle Swarm Optimization,” J.

Ocean Univ. China, vol. 21, no. 1, pp. 28–38, 2022, doi:

10.1007/s11802-022-4709-8.

[26] R. Dębski and R. Dreżewski, “Multi-Objective Ship

Route Optimisation Using Estimation of Distribution

Algorithm,” Appl. Sci., vol. 14, no. 13, 2024, doi:

10.3390/app14135919.

[27] X. Li, H. Wang, and Q. Wu, “Multi-objective

optimization in ship weather routing,” 2017 Constr.

Nonsmooth Anal. Relat. Top. (Dedicated to Mem. V.F.

Demyanov), CNSA 2017 - Proc., pp. 1–4, 2017, doi:

10.1109/CNSA.2017.7973982.

[28] R. Szlapczynski, J. Szlapczynska, and R. Vettor, “Ship

weather routing featuring w-MOEA/D and uncertainty

handling,” Appl. Soft Comput., vol. 138, p. 110142, 2023,

doi: 10.1016/j.asoc.2023.110142.

[29] R. Szlapczynski and J. Szlapczynska, “W-dominance:

Tradeoff-inspired dominance relation for preference-

based evolutionary multi-objective optimization,”

Swarm Evol. Comput., vol. 63, no. March 2020, p. 100866,

2021, doi: 10.1016/j.swevo.2021.100866.

103

[30] Y. Chen, “Department of Mechanics and Maritime

Sciences Isochrone-Based Voyage Optimization Methods

to Increase Shipping Energy Efficiency,” 2024.

[31] Y. W. Shin et al., “Near-optimal weather routing by using

improved A* algorithm,” Appl. Sci., vol. 10, no. 17, 2020,

doi: 10.3390/app10176010.

[32] L. Qian, H. Li, M. Hong, Y. Qi, and Z. Guo, “Avoiding

sudden maritime risk: A new variable speed route

planning model by integrating Spatio-temporal

dimension,” Ocean Eng., vol. 288, no. P1, p. 115950, 2023,

doi: 10.1016/j.oceaneng.2023.115950.

[33] D. Ma, S. Zhou, Y. Han, W. Ma, and H. Huang, “Multi-

objective ship weather routing method based on the

improved NSGA-III algorithm,” J. Ind. Inf. Integr., vol. 38,

no. January, p. 100570, 2024, doi: 10.1016/j.jii.2024.100570.

[34] G. Mannarini, D. N. Subramani, P. F. J. Lermusiaux, and

N. Pinardi, “Graph-Search and Differential Equations for

Time-Optimal Vessel Route Planning in Dynamic Ocean

Waves,” IEEE Trans. Intell. Transp. Syst., vol. 21, no. 8,

pp. 3581–3593, 2020, doi: 10.1109/TITS.2019.2935614.

[35] G. Mannarini, M. L. Salinas, L. Carelli, N. Petacco, and J.

Orović, “VISIR-2: Ship weather routing in Python,”

Geosci. Model Dev., vol. 17, no. 10, pp. 4355–4382, 2024,

doi: 10.5194/gmd-17-4355-2024.

[36] F. Yin and F. Shi, A Comparative Survey of Big Data

Computing and HPC: From a Parallel Programming

Model to a Cluster Architecture, vol. 50, no. 1. Springer

US, 2022. doi: 10.1007/s10766-021-00717-y.

[37] D. K. Panda, Ed., Supercomputing Frontiers, vol. 12082.

in Lecture Notes in Computer Science, vol. 12082. Cham:

Springer International Publishing, 2020. doi: 10.1007/978-

3-030-48842-0.

[38] F. Sattler, B. Carrillo-Perez, S. Barnes, K. Stebner, M.

Stephan, and G. Lux, “Embedded 3D reconstruction of

dynamic objects in real time for maritime situational

awareness pictures,” Vis. Comput., vol. 40, no. 2, pp. 571–

584, 2024, doi: 10.1007/s00371-023-02802-4.

[39] T. Nakaegawa, “High-Performance Computing in

Meteorology under a Context of an Era of Graphical

Processing Units,” Computers, vol. 11, no. 7, 2022, doi:

10.3390/computers11070114.

[40] M. Liang, K. Liu, R. Gao, and Y. Li, “Integrating GPU-

Accelerated for Fast Large-Scale Vessel Trajectories

Visualization in Maritime IoT Systems,” IEEE Trans.

Intell. Transp. Syst., vol. 26, no. 3, pp. 4048–4065, 2025,

doi: 10.1109/TITS.2024.3521050.

[41] E. Alhenawi, R. A. Khurma, A. A. Sharieh, O. Al-Adwan,

A. Al Shorman, and F. Shannaq, “Parallel Ant Colony

Optimization Algorithm for Finding the Shortest Path for

Mountain Climbing,” IEEE Access, vol. 11, no. January,

pp. 6185–6196, 2023, doi: 10.1109/ACCESS.2022.3233786.

[42] S. A. M. Ali and E. H. Al-Hemiary, “A Parallel

Implementation Method for Solving the Shortest Path

Problem for Vehicular Networks,” Proc. 2020 1st Inf.

Technol. to Enhanc. E-Learning other Appl. Conf. IT-ELA

2020, pp. 121–126, 2020, doi: 10.1109/IT-

ELA50150.2020.9253086.

[43] Z. Cheng, L. Zhao, and Z. Shi, “Decentralized Multi-

UAV Path Planning Based on Two-Layer Coordinative

Framework for Formation Rendezvous,” IEEE Access,

vol. 10, pp. 45695–45708, 2022, doi:

10.1109/ACCESS.2022.3170583.

[44] N. Edmonds, A. Breuer, D. Gregor, and A. Lumsdaine,

“Single-source shortest paths with the parallel boost

graph library,” no. May 2014, pp. 219–248, 2009, doi:

10.1090/dimacs/074/09.

[45] J. Kim, “HGED: A Hybrid Search Algorithm for Efficient

Parallel Graph Edit Distance Computation,” IEEE Access,

vol. 8, pp. 175776–175787, 2020, doi:

10.1109/ACCESS.2020.3026648.

[46] D. Josifoski, M. Gusev, V. Zdraveski, and S. Ristov,

“Parallelization of Dijkstra’s Algorithm for Robot Motion

Planning: Is It Worth Increasing Speed Without Losing

Accuracy?,” 2022 30th Telecommun. Forum, TELFOR

2022 - Proc., pp. 1–4, 2022, doi:

10.1109/TELFOR56187.2022.9983755.

[47] J. H. Seo, H. Lee, and K. D. Kim, “A Parallelization

Algorithm for Real-Time Path Shortening of High-DOFs

Manipulator,” IEEE Access, vol. 9, pp. 123727–123741,

2021, doi: 10.1109/ACCESS.2021.3109744.

[48] Y. Chen et al., “State-of-the-art optimization algorithms

in weather routing — ship decision support systems:

challenge, taxonomy, and review,” Ocean Eng., vol. 331,

no. April, p. 121198, 2025, doi:

10.1016/j.oceaneng.2025.121198.