Enhancing Many-Objective Particle Swarm Optimization with

Island Model for Agricultural Optimization

Chnini Samia

, Abadlia Houda

, Smairi Nadia

and Nasri Nejah

SETIT Laboratory, Faculty of Sciences of Gafsa, University of Gafsa, Tunisia

Univ. Manouba, ENSI, LARIA UR22ES01, Campus Universitaire Manouba, Tunisia

Nadia Smairi, COSMOS Laboratory, National School of Computer Sciences, University of Manouba, Tunisia

Keywords: Many-Objective Optimization, Distributed Optimization, Island Model.

Abstract: With the growing complexity of agricultural systems and the need to optimize multiple conflicting objectives

simultaneously, traditional optimization methods often struggle to find satisfactory solutions. In this work,

we introduce a novel enhancement to the standard Multi Objectives Particle Swarm Optimization (MOPSO)

algorithm that significantly improves its effectiveness in handling the diverse and dynamic objectives inherent

in agricultural optimization problems. we propose an improvement to the MOPSO algorithm by introducing

an islanding technique to promote exploration and exploitation of the many-objective search space. The

improved MOPSO algorithm, called I-MOPSO guide the search towards optimal and diverse solutions by

dividing the search space into islands and facilitating information exchange between them. We put I-MOPSO

into practice and tested it using a series of common many objective optimization algorithms. According to

Experimental results show that I-MOPSO is capable of finding high-quality solutions on a variety of test

problems, often outperforming the standard MOPSO algorithm and NSGAIII.

1 INTRODUCTION

Numerous many-objective optimization problems

(MaOOPs) are encountered in many scientific and

engineering study domains. In contemporary

agriculture, the optimization of multiple conflicting

objectives has become increasingly vital for

sustainable and efficient agricultural practices.

Farmers and agricultural planners are confronted with

complex decision-making scenarios involving trade-

offs between maximizing crop yield, enhancing

resource utilization efficiency, and ensuring

environmental sustainability (Anosri et al., 2023; Liu,

Shen, Yang, & Yang, 2013)

When dealing with many-objective optimization

problems (MaOPs), optimization issues with four or

more competing objectives, certain traditional

MOEAs struggle with diversity and convergence,

despite their effectiveness for two or three-objective

problems.

https://orcid.org/0000-0003-2334-2193

https://orcid.org/0000-0003-3467-8058

https://orcid.org/0000-0001-5568-8832

https://orcid.org/0000-0002-9293-8383

Nonetheless, a wide variety of research indicates

that evolutionary multi-objective optimization

(EMOO) algorithms are unable to address MaOO

issue. (Rakshit, Chowdhury, Konar, & Nagar, 2020)

The study presented is part of larger context of

precision agriculture, an approach that integrates

technological advances to optimize farming practices.

The current metaheuristic algorithms still have a

number of shortcomings despite the specific

advances, including a slow convergence rate, a

tendency to trap in local optima, the use of

complicated operators, lengthy computation times. In

particular, they encounter problems of premature

convergence in the case of multimodal and high-

dimensional problems. Furthermore, current

knowledge indicates that bio-inspired and meta-

heuristic algorithms do not always achieve the

required performance levels. Their computation time

can vary according to the complexity of the problem

and the nature of the solution sought. So, some

608

Samia, C., Houda, A., Nadia, S. and Nejah, N.

Enhancing Many-Objective Particle Swarm Optimization with Island Model for Agricultural Optimization.

DOI: 10.5220/0013315200003890

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 17th International Conference on Agents and Artiﬁcial Intelligence (ICAART 2025) - Volume 1, pages 608-615

ISBN: 978-989-758-737-5; ISSN: 2184-433X

researchers have explored architecture modifications,

parameter adjustments and partial operator

improvements to overcome these weaknesses. In

conclusion, our study seeks to harness the potential of

advanced technologies, in particular bio-inspired

algorithms, to solve the complex challenges of

modern agriculture (Maraveas et al., 2023).

2 METHODOLOGY

2.1 MOPSO

Numerous research has shown the effectiveness of

MOPSO in finding optimal solutions for a wide range

of real-world problems. (Al-Hassan, Fayek, &

Shaheen, 2006)

In (Yang, Tang, Cai, Chen, & Hu, 2022) the

authors propose a new cooperative framework with

double elite selection and one-dimensional chaotic

logistic perturbation (LCSDP), which leads to a

considerable increase in convergence and diversity of

solutions. Each class performs internal migrations to

explore the search space and optimize the solutions in

that class. thus, subpopulations exchange information

and promote solution diversity. In this way, by using

a diversity-based selection technique, the authors are

able to prevent an early convergence to a single local

optimum. By combining an island model with a local

search method (Variable Neighborhood Search). In

(Abadlia, Smairi, & Ghedira, 2017, 2018) the authors

suggests an enhancement of the MOPSO algorithm

that strikes a balance between search space

exploration and exploitation. The approach uses local

search to maximize local solutions on each island

while combining dynamic exchanges and

subpopulation movement (islands) to preserve

variety.

2.2 Island Model

Island models divide the population into numerous

subpopulations known as islands in order to

parallelize the evolution process (Wu, Mallipeddi, &

Suganthan, 2019). The subpopulations periodically

exchange solutions between islands through a process

called migration. This migration process plays a

crucial role in maintaining the diversity of the islands

(Wu et al., 2019; Delaram Yazdani et al., 2023).

The result is a dynamic and hardy population that

can react to fresh chances and challenges as a group,

guaranteeing the long-term survival of the sub

population (Khediri, Nasri, Khan, & Kachouri, 2021).

The idea of changing population size has been applied

in a number of evolutionary computation subfields,

however its function varies depending on the context

(Al-Hassan et al., 2006; Danial Yazdani, Omidvar,

Branke, Nguyen, & Yao, 2019).

Every particle is contained within a subswarm of

an island, and the topology of the island determines

its neighborhoods. Every subpopulation comprises an

equal number of individuals and only optimizes a

single objective (Yang et al., 2022)(Chnini, Smairi, &

Nasri, 2024).

2.3 The Structure and Planning of

Islands Topology

Figure 1: Island topology, a.ring topology, b.star topology,

c. grid topology.

The distribution of islands can be designed in many

ways according to the topology chosen, such as ring,

star or grid patterns. (J. Li & Gonsalves, 2022)

Particles moving between islands usually takes

place in a periodic manner and follows a set of rules

of migration. This is done to promote diversity and

reduce the chances of premature convergence.

(Baltazar, 2015; Rakshit et al., 2020) (J. Li &

Gonsalves, 2022; Delaram Yazdani et al., 2023).

3 RELATED WORK

The ability of MOPSO algorithm to explore Pareto

fronts efficiently and enhance solution convergence

has led to various adaptations and improvements

tailored to the specific requirements of each

application (Anosri et al., 2023).

In (Jithendranath & Das, 2023), authors proposed

a new variant of the MOPSO algorithm called NLTV-

MOPSO to optimize power flow in island-mode

microgrids with photovoltaic generation.

In (H. Li, Wang, Lan, Wu, & Zeng, 2023), the

authors proposed a new dynamic multi-objective

optimization algorithm based on non-inductive

Enhancing Many-Objective Particle Swarm Optimization with Island Model for Agricultural Optimization

609

transfer learning, called MSAS-DMOA (Multi-

Strategy Adaptive Selection-DMOA). This algorithm

uses a combination of adaptive strategies to solve

dynamic multi-objective optimization problems

(DMOPs). By integrating transfer learning and the

MOPSO algorithm, it improves convergence and

solution diversity.

A better MOPSO algorithm, known as f-MOPSO-

II, has been proposed in (Mansouri, Safavi, & Rezaei,

2022) to optimize reservoir management in the

context of climate change.

In (Zarei, Azari, & Heidari, 2022) improved the

performance of the NSGA-II and MOPSO algorithms

for multi-objective optimization of urban water

distribution networks by modifying the decision

space. The results showed that modifying the decision

space, with the integration of the penalty for

exceeding authorized pressure limits.

The authors in (Reddy & Kumar, 2009) have

developed an algorithm called EM-MOPSO (Elitist

Mutated Multi-Objective Particle Swarm

Optimization) for integrated water resource

management. Their approach combines elitist and

mutation mechanisms to improve solution diversity

and accelerate convergence.

4 PROBLEM DEFINITION

A detailed description of the objective functions used in our

agricultural optimization study is presented.

4.1 The Optimization Function for

Allocating Water

This objective function seeks to minimize the

absolute difference between water supply (supplied

by sources) and water demand for each site and at

each time step. (Nouiri, Yitayew, Maßmann, &

Tarhouni, 2015).

𝑓



=𝑀𝑎𝑥 𝐹𝐷

(

𝑠𝑒,𝑑,𝑡



−

𝐷

(

𝑑,𝑡



𝐷

(

𝑑,𝑡











𝑑=1,…,𝑁𝐷



𝑎𝑛𝑑 𝑇



,…𝑇



(1)

4.2 Target Function for Drawdown

Reduction

This function assesses how much the exploitation of

an aquifer reduces relative drawdowns. (Nouiri,

Yitayew, Maßmann, & Tarhouni, 2015).

𝐹



(

𝑠𝑒,𝑑,𝑡



<𝐹



𝐷

(

𝑠𝑒, 𝑑



∀𝑠𝑒, ∀𝑑 𝑎𝑛𝑑 ∀𝑡

(2)

4.3 Objective Function with Penalty for

Maximum Permitted Infraction for

Pumping

This function calculates the discrepancy between the

maximum pumping rates that have been set and the

water flows that the wells have extracted.

𝑓



(

𝑠



=𝑀𝑎𝑥(

𝐷𝑊𝐷𝑊

(

𝑤,𝑡



×𝑉𝑀𝑎𝑥𝐷𝐷

(

𝑤



𝐻𝑖

(

𝑤



− 𝐵𝑂𝑇𝑀

(

𝑤



×1

(

𝑤



𝑤=1,…,𝑁



𝑎𝑛𝑑𝑡=𝑇



,…𝑇



(3)

4.4 Objective Function with Penalty for

Exceeding Maximum Acceptable

Drawdown

The percentage of the maximum allowable drawdown

that is exceeded determines the penalty term. These

solutions become less appealing if the drawdown over

the threshold because the penalty term raises the

value of the objective function.

The problem of water distribution is a complex

and many objective problem requires efficient

solutions to satisfy often contradictory requirements

(Nouiri, Yitayew, Maßmann, & Tarhouni, 2015).

𝑉𝑀𝑎𝑥𝐷𝐷

(

𝑤



=𝑀𝑎𝑥(

𝐷𝑊𝐷𝑊

(

𝑤,𝑡



𝐷𝐷



(

𝑤



𝑤=1,…𝑁



𝑎𝑛𝑑 𝑡=𝑇



,𝑇



(4)

5 APPROACH

The MOPSO algorithm has proven its effectiveness

in reesolving multi-objective problems. However,

like many optimization algorithms, it has certain

limitations, most notably premature convergence and

low diversity in the solutions proposed. In order to

improve the quality of solutions and avoid local

minima, several adjustments and improvements have

been proposed in this work. These seek to improve

diversity, optimize exploration, and dynamically

modify the algorithm's settings.

Our optimization strategy incorporates the

previously mentioned objective functions through the

utilization of the MOPSO algorithm, which has been

augmented by the island model. In this study, we have

selected the ring and star topologies to examine how

their interaction dynamics influence solution

convergence and diversity.

The I-MOPSO algorithm begins by dividing the

global population into several sub-populations, called

ICAART 2025 - 17th International Conference on Agents and Artiﬁcial Intelligence

610

islands. Each island contains a number of particles,

representing potential solutions to the optimization

problem. Each algorithm had a population size of 100

particles or solutions, and the maximum number of

iterations was set at 600.The particles on each island

are randomly initialized in the space of decision

variables, and their velocity is set randomly. Each

particle has a memory of the best solution it has

encountered so far, called bestPosition, as well as an

archive of non-dominated solutions.Over the course

of iterations, particles update their position in the

search space. This update is influenced by three

components : inertia (which retains part of the

previous velocity), a cognitive term (influenced by

the best individual position reached by the particle),

and a social term (influenced by a leader selected

from the island's archive of non-dominated

solutions). Periodically, information is exchanged

between islands (each 20 iterations). For I-MOPSO,

the star topology is used to connect the islands.

Islands share their best solutions to date. A percentage

of the best particles on each island migrate to the other

islands. This process promotes solution diversity by

allowing different sub-populations to exchange

information. After each exchange phase, the overall

best solution is updated, taking into account the best

solutions from all the islands.This island approach

enables a more complete exploration of the search

space. Each island explores a part of this space

independently, thus increasing the diversity of

solutions explored.

5.1 Experiments and Metrics

In order to assess the effectiveness of the proposed I-

MOPSO method in a star topology, we created a

series of tests to investigate how important

parameters affect the caliber of the results. The

number of islands, the number of particles per island

and the migration rate are the relevant parameters. We

compare the performance of our I-MOPSO model,

NSGA-III and MOPSO in a many-objective

optimization problem, specifically applied to water

distribution management. Finding out how well these

algorithms perform in terms of diversity of generated

solutions and convergence to the Pareto front is the

aim. We changed the number of islands (3, 5, 10), the

number of particles per island (100, 200, 300) and the

migrate rates (0.1,0.5,0.8) in order to do this. These

factors have a significant impact on the quality of the

solutions and the rate of convergence of population-

based algorithms.

• Performance Metrics

Two indicators are used in this paper to assess the

algorithm.On the one hand, the method is evaluated

using the coverage metric known as the C metric is

a performance indicator used to assess the degree of

coverage between two solution sets produced during

optimization. (Selvam, Vinod Kumar, & Siripuram,

2017). This quality indicator can be assessed in the

manner described below.

𝐶=



𝑦∈𝑃



∃𝑋∈𝑃



;𝑥>𝑦



𝑃



(5)

If 𝐶

(

𝑃



,𝑃





< 𝐶

(

𝑃



,𝑃





, the pareto front P

have

the better solutions than P

(Selvam, Vinod

Kumar, & Siripuram, 2017)

In the other hand, hypervolume (J. Li &

Gonsalves, 2022; Delaram Yazdani et al., 2023) is

also algorithm’s evaluation indication. Additionally,

it provides a through indication for assessing

distribution and convergence. A higher hypervolume

reflects better diversity of solutions.

5.2 Results and Discussion

The tables 1, 2 and 3 displays the C metric findings

for comparisons between I-MOPSO, NSGA-III and

MOPSO as a function of the following algorithm

parameters: population size, migration rate, and

number of islands. The C measure shows the

proportion of solutions on one front that are

dominated or covered by those on another front.

Regarding the comparison of MOPSO and I-

MOPSO, the latter appears to encompass a significant

portion of the MOPSO front. As soon as the

population size reaches 200 or more, I-MOPSO gains

complete dominance over NSGA-III when the

number of islands is set to 3.

Furthermore, NSGA-III is totally overpowered by

MOPSO in these configurations, suggesting that

NSGA-III performs worse in these

circumstances.According to the C metric values, I-

MOPSO continuously outperforms NSGA-III across

all configurations, attaining 100% coverage for a

number of parameter combinations. This indicates

that I-MOPSO frequently outperforms MOPSO,

especially when the number of islands and population

values are high.

Enhancing Many-Objective Particle Swarm Optimization with Island Model for Agricultural Optimization

611

Table 1: Performance of the algorithms using C metric (MigRate=0.1).

Mig

Rate

C(IMOPSO,

SGAIII)

C(NSGAIII,

IMOPSO)

C(MOPSO,

SGAIII)

C(NSGAIII,

MOPSO)

C(MOPSO,

IMOPSO

C(IMOPSO

,MOPSO

3 0.1

100 100 0 100 0 86,96 99,58

200 100 0 100 0 99,17 97,19

300 100 0,30 100 0 84,05 98,79

5 0.1

100 100 0,36 100 0 88,85 99,61

200 100 0 100 0 86,98 97,17

300 100 0 100 0 95,3 95,85

10 0.1

100 100 0 100 0 80,65 99,59

200 100 0 100 0 97,07 98,72

300 100 0 100 0 97,59 99,29

Table 2: Performance of the algorithms using C metric (MigRate=0.5).

MigRa

Pop

C(IMOP,N

SGAIII)

C(NSGAIII,I

-MOPS)

C(MOPSO,

SGAIII)

C(NSGAIII,M

OPSO)

C(MOPSO,I

-MOPSO

C(IMOPSO,

MOPSO)

3 0.5

100 100 0 100 1,73 96,54 91,47

200 100 0 100 0 93,74 96,45

300 100 0 100 0 67,22 99,97

5 0.5

100 100 0 100 0 99,94 65,65

200 100 0 100 0,36 84,75 99,27

300 100 0 100 1,02 86,77 99,33

10 0.5

100 100 0 100 0 94,81 99,47

200 100 0 100 0 96,89 98,63

300 100 0 100 0 74,13 99,91

Table 3: Performance of the algorithms using C metric (MigRate=0.8).

MigRa

Pop

C(IMOP,N

SGAIII)

C(NSGAIII,I

-MOPS)

C(MOPSO,N

SGAIII)

C(NSGAIII,

MOPSO)

C(MOPSO,I

-MOPSO

C(IMOPSO,

MOPSO)

3 0.8

100 100 0 100 0 90,05 97,44

200 100 0 100 0 90,46 98,99

300 100 0 100 0 99,84 64,25

5 0.8

100 100 0 100 0 99,31 80,33

200 100 0 100 0,56 99,66 93,68

300 100 0 100 1,21 87,76 99,66

10 0.8

100 100 0 100 0 98,86 91,32

200 100 0 100 0 96,25 99,41

300 100 0 100 0 98,51 96,06

5.2.1 Impact of the Number of Islands:

Number of Islands

(Numislands = 3, 5, 10)

The number of islands is a determining factor in the

effectiveness of I-MOPSO. With only three islands,

I-MOPSO dominates NSGA-III but shows partial

coverage over MOPSO. However, I-MOPSO’s

dominance over MOPSO increases with the number

of particles, indicating that even with a small number

of islands, I-MOPSO can be competitive if the

population size is sufficient. With 10 islands, I-

MOPSO reaches its maximum performance,

completely dominating NSGA-III and MOPSO in

several configurations. Then, a large number of

islands gave I-MOPSO increased capacity to

effectively explore the Pareto front.

5.2.2 Impact of Number of Particles:

(Particles = 100, 200, 300)

With a population of 100, I-MOPSO manages to

dominate NSGA-III but shows partial coverage of

MOPSO's Pareto front, particularly when the number

of islands is minimal. By increasing the population to

200, I-MOPSO improves its coverage, achieving

ICAART 2025 - 17th International Conference on Agents and Artiﬁcial Intelligence

612

complete domination of NSGA-III and superior

coverage than MOPSO. This suggests that a

population of 200 is sufficient to enable I-MOPSO to

generate a high-quality Pareto front in moderate

migration and island number configurations.

5.2.3 Impact of Migration Rate

(MigrationRate=0.1, 0.5,0.8)

With a migration rate of 0.8, I-MOPSO achieves high

dominance results, indicating that the frequent

exchange of individuals between islands improves the

diffusion of optimal solutions. This high migration

rate promotes rapid convergence towards the Pareto

front while maintaining a diversity of solutions,

which is essential for the quality of the front

generated. A lower migration rate might retain more

local diversity, but could slow convergence.

A migration rate of 0.8 appears to be the best

setting for I-MOPSO overall, providing for a balance

between Pareto front exploration and exploitation.

Analyzing the findings reveals that I-

MOPSO performs the best in most configurations,

particularly when paired with a big population (200–

300), a migration rate of 0.8, and a high number of

islands.

In order to evaluate the I-MOPSO algorithm's

performance, we measured the mean hypervolume

(Mean Hypervolume) and the standard deviation of

hypervolume (Std Hypervolume) throughout the last

50 iterations of each execution. Comparative analysis

of the obtained hypervolumes provides essential

information on how parameters affect the algorithm's

ability to efficiently search the space of solutions while

maintaining a steady convergence towards the Pareto

front. These findings make it possible to determine the

best configurations for many objective problems in

order to maximize solution diversity and convergence.

Figure 2: Comparison of hypervolume evolution for over

iterations.

The figure 2 illustrates differences in the stability

and efficiency of algorithms for maximizing

hypervolume over the course of iterations. I-MOPSO

seems to indicate higher variability, with multiple

notable peaks, which would suggest a more intensive

search for solutions. Reflecting a more gradual

convergence, NSGA-III exhibits fewer fluctuation and

is comparatively stable. In terms of hypervolume

values, MOPSO exhibits regular fluctuations but is still

generally less effective than I-MOPSO.

By obtaining a greater mean hypervolume, I-

MOPSO continuously surpasses NSGA-III and

MOPSO, demonstrating its superior exploration and

exploitation capabilities. whereas NSGA-

III consistently achieves the lowest mean

hypervolume. I-MOPSO maintains its edge in the

majority of cases (table 4,5 and 6), whereas NSGA-

III becomes more competitive as migration rates rise.

Both I-MOPSO and MOPSO's mean

hypervolume improves with population size, with I-

MOPSO maintaining a slight advantage.

NSGAIII continues to be the least competitive in

mean hypervolume and stability, despite a minor

improvement with bigger populations.

Table 4: Means ans stds Hypervolume values throughout the last 50 iterations (MigRate=0.1).

umI MigRate Pop

I-MOPSO

SGAIII MOPSO

Mean

Hypervolume

Std

Hypervolume

Mean

Hypervolume

Std

Hypervolume

Mean

Hypervolume

Std

Hypervolume

3 0.1

100 1.223e+00 8.147e-01 6.029e-01 1.124e-01 1.109e+00 6.185e-01

200 1.256e+00 9.117e-01 6.321e-01 1.324e-01 1.171e+00 7.330e-01

300 2.106e+00 1.768e+00 6.290e-01 1.263e-01 1.554e+0 1.020e+00

5 0.1

100 1.117e+00 7.743e-01 6.270e-01 1.081e-01 9.974e-01 6.410e-01

200 1.501e+00 9.620e-01 6.473e-01 1.251e-01 1.472e+00 1.016e+00

300 1.725e+00 1.419e+00 6.188e-01 1.208e-01 1.494e+00 1.087e+00

10 0.1

100 1.313e+00 6.939e-01 6.403e-01 1.447e-01 1.062e+00 8.615e-01

200 1.867e+00 1.356e+00 6.125e-01 1.274e-01 1.400e+00 7.621e-01

300 2.393e+00 1.770e+00 5.913e-01 1.070e-01 1.459e+00 9.708e-01

Enhancing Many-Objective Particle Swarm Optimization with Island Model for Agricultural Optimization

613

Table 5: Means ans stds Hypervolume values throughout the last 50 iterations(MigRate=0.5).

NumI MigRate Pop

I-MOPSO

SGAIII MOPSO

Mean

Hypervolume

Std

Hypervolume

Mean

Hypervolume

Std

Hypervolume

Mean

Hypervolume

Std

Hypervolume

3 0.5

100 1.163e+00 5.071e-01 6.250e-01 1.186e-01 1.210e+00 6.833e-01

200 1.781e+00 1.349e+00 6.120e-01 1.000e-01 1.411e+00 8.732e-01

300 1.772e+00 1.299e+00 5.953e-01 1.246e-01 1.578e+00 1.023e+00

5 0.5

100 9.974e-01 6.986e-01 6.438e-01 1.461e-01 9.559e-01 4.154e-01

200 1.745e+00 1.281e+00 6.104e-01 1.206e-01 1.476e+00 9.575e-01

300 1.781e+00 1.191e+00 6.105e-01 1.117e-01 1.511e+00 1.002e+00

10 0.5

100 1.290e+00 8.168e-01 6.376e-01 1.225e-01 1.149e+00 5.246e-01

200 2.269e+00 1.862e+00 6.040e-01 1.153e-01 1.140e+00 5.567e-01

300 2.353e+00 1.957e+00 6.174e-01 1.264e-01 1.315e+00 8.303e-01

Table 6: Means ans stds Hypervolume values throughout the last 50 iterations (MigRate=0.8).

umI MigRate Pop

I-MOPSO

SGAIII MOPSO

Mean

Hypervolume

Std

Hypervolume

Mean

Hypervolume

Std

Hypervolume

Mean

Hypervolume

Std

Hypervolume

3 0.8

100 1.092e+00 5.180e-01 6.267e-01 1.150e-01 1.254e+00 7.429e-01

200 1.314e+00 8.915e-01 6.463e-01 1.200e-01 1.255e+00 9.626e-01

300 1.975e+00 1.700e+00 6.481e-01 1.320e-01 1.889e+00 1.404e+00

5 0.8

100 1.175e+00 7.390e-01 6.455e-01 1.243e-01 1.133e+00 6.587e-01

200 1.789e+00 1.741e+00 6.707e-01 1.688e-01 1.420e+00 8.965e-01

300 1.635e+00 1.001e+00 6.097e-01 1.261e-01 1.428e+00 9.532e-01

10 0.8

100 1.393e+00 9.896e-01 6.240e-01 1.340e-01 9.884e-01 4.672e-01

200 1.501e+00 1.061e+00 6.146e-01 1.033e-01 1.601e+00 1.135e+00

300 2.760e+00 1.763e+00 6.066e-01 1.202e-01 1.345e+00 1.097e+00

6 CONCLUSIONS

In this study, we have proposed an innovative many-

objective approach based on the Island model to solve

the water distribution optimization problem, taking

into account several conflicting objectives. The

distinguishing feature of this method is its

decomposition of the problem into several

subpopulations spread over islands, enabling more

efficient exploration of the solution space by

combining local search and migration strategies

between islands.

Our experiments on agricultural optimization

problems show that this method can effectively find

many high-quality solutions. When compared to other

evolutionary algorithms, it does better at solving

agricultural problems with many objectives. This study

improves optimization techniques for agriculture and

opens up new avenues for future research in this area.

Finally, our analysis shows that the adjustment of

parameters such as the number of islands and particles

plays an essential role in improving the performance of

our model. The proposed approach thus offers a robust

and flexible method, which can be adapted to a wide

variety of many-objective problems in the field of

resource optimization.

REFERENCES

Abadlia, H., Smairi, N., & Ghedira, K. (2017). Particle

swarm optimization based on dynamic island model.

Paper presented at the 2017 IEEE 29th International

Conference on Tools with Artificial Intelligence

(ICTAI).

Abadlia, H., Smairi, N., & Ghedira, K. (2018). A hybrid

Immigrants schema for particle swarm optimization

algorithm. Procedia Computer Science, 126, 105-115.

Al-Hassan, W., Fayek, M., & Shaheen, S. (2006). Psosa:

An optimized particle swarm technique for solving the

urban planning problem. Paper presented at the 2006

international conference on computer engineering and

systems.

Anosri, S., Panagant, N., Champasak, P., Bureerat, S.,

Thipyopas, C., Kumar, S., . . . Yildiz, A. R. (2023). A

comparative study of state-of-the-art metaheuristics for

solving many-objective optimization problems of fixed

wing unmanned aerial vehicle conceptual design.

ICAART 2025 - 17th International Conference on Agents and Artiﬁcial Intelligence

614

Archives of Computational Methods in Engineering,

30(6), 3657-3671.

Baltazar, S. M. C. (2015). A Study on the Deployment of GA

in a Grid Computing Framework. Universidade do

Algarve (Portugal),

Chnini, S., Smairi, N., & Nasri, N. (2024). Classification

survey of many-objective optimization methods. Paper

presented at the 2024 10th International Conference on

Control, Decision and Information Technologies

(CoDIT).

Jithendranath, J., & Das, D. (2023). Multi-objective optimal

power flow in islanded microgrids with solar PV

generation by NLTV-MOPSO. IETE Journal of

Research, 69(4), 2130-2143.

Khediri, S. E., Nasri, N., Khan, R. U., & Kachouri, A.

(2021). An improved energy efficient clustering

protocol for increasing the life time of wireless sensor

networks. Wireless Personal Communications, 116,

539-558.

Li, H., Wang, Z., Lan, C., Wu, P., & Zeng, N. (2023). A

novel dynamic multiobjective optimization algorithm

with non-inductive transfer learning based on multi-

strategy adaptive selection. IEEE Transactions on

Neural Networks and Learning Systems.

Li, J., & Gonsalves, T. (2022). Parallel hybrid island

metaheuristic algorithm. IEEE Access, 10, 42268-

42286.

Liu, Y., Shen, H., Yang, W., & Yang, J. (2013).

Optimization of agricultural BMPs using a parallel

computing based multi-objective optimization

algorithm. Environmental Resources Research, 1(1),

39-50.

Mansouri, M., Safavi, H. R., & Rezaei, F. (2022). An

improved MOPSO algorithm for multi-objective

optimization of reservoir operation under climate

change. Environmental Monitoring and Assessment,

194(4), 261.

Maraveas, C., Asteris, P. G., Arvanitis, K. G., Bartzanas,

T., & Loukatos, D. (2023). Application of bio and

nature-inspired algorithms in agricultural engineering.

Archives of Computational Methods in Engineering,

30(3), 1979-2012.

Muthukumaran, S., Geetha, P., & Ramaraj, E. (2023).

Multi-Objective Optimization with Artificial Neural

Network Based Robust Paddy Yield Prediction Model.

Intelligent Automation & Soft Computing, 35(1).

Nouiri, I. Outils d’aide à la décision pour la gestion

optimale des ressources en eau.

Nouiri, I., Yitayew, M., Maßmann, J., & Tarhouni, J.

(2015). Multi-objective optimization tool for integrated

groundwater management. Water Resources

Management, 29, 5353-5375.

Rakshit, P., Chowdhury, A., Konar, A., & Nagar, A. K.

(2020). Migration in multi-population differential

evolution for many objective optimization. Paper

presented at the 2020 IEEE Congress on Evolutionary

Computation (CEC).

Reddy, M. J., & Kumar, D. N. (2009). Performance

evaluation of elitist-mutated multi-objective particle

swarm optimization for integrated water resources

management. Journal of Hydroinformatics, 11(1), 79-88.

Selvam, K., Vinod Kumar, D., & Siripuram, R. (2017).

Distributed generation planning using peer enhanced

multi-objective teaching–learning based optimization

in distribution networks. Journal of The Institution of

Engineers (India): Series B, 98

, 203-211.

Sivagurunathan, G., Kotteeswaran, R., Suresh, M., &

Kirthini Godweena, A. (2021). Design of centralized

controller for multivariable process using MOPSO

algorithm. Indian Journal of Science and Technology,

14(26), 2223-2237.

Wu, G., Mallipeddi, R., & Suganthan, P. N. (2019).

Ensemble strategies for population-based optimization

algorithms–A survey. Swarm and Evolutionary

Computation, 44, 695-711.

Yang, N., Tang, Z., Cai, X., Chen, L., & Hu, Q. (2022).

Cooperative multi-population Harris Hawks

optimization for many-objective optimization. Complex

& Intelligent Systems, 8(4), 3299-3332.

Yazdani, D., Cheng, R., He, C., & Branke, J. (2020).

Adaptive control of subpopulations in evolutionary

dynamic optimization. IEEE transactions on

cybernetics, 52(7), 6476-6489.

Yazdani, D., Omidvar, M. N., Branke, J., Nguyen, T. T., &

Yao, X. (2019). Scaling up dynamic optimization

problems: A divide-and-conquer approach. IEEE

transactions on evolutionary computation, 24(1), 1-15.

Yazdani, D., Yazdani, D., Yazdani, D., Omidvar, M. N.,

Gandomi, A. H., & Yao, X. (2023). A species-based

particle swarm optimization with adaptive population

size and deactivation of species for dynamic

optimization problems. ACM Transactions on

Evolutionary Learning and Optimization, 3(4), 1-25.

Zarei, N., Azari, A., & Heidari, M. M. (2022). Improvement

of the performance of NSGA-II and MOPSO

algorithms in multi-objective optimization of urban

water distribution networks based on modification of

decision space. Applied Water Science, 12(6), 133.

Enhancing Many-Objective Particle Swarm Optimization with Island Model for Agricultural Optimization

615