See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/388422560

Optimizing LPB Algorithms using Simulated Annealing

Preprint · December 2024

DOI: 10.48550/arXiv.2501.14751

CITATIONS READS
0 7

2 authors:

Dana Hamad Tarik A. Rashid

Soran University University of Kurdistan Hewler
11 PUBLICATIONS 3 CITATIONS 519 PUBLICATIONS 12,297 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Dana Hamad on 30 January 2025.

The user has requested enhancement of the downloaded file.

Optimizing LPB Algorithms using Simulated Annealing

Dana Rasul Hamad1. Tarik A. Rashid2*

Abstract. Learner Performance-based Behavior using Simulated Annealing (LPBSA) is an

improvement of the Learner Performance-based Behavior (LPB) algorithm. LPBSA, like LPB, has
been proven to deal with single and complex problems. Simulated Annealing (SA) has been
utilized as a powerful technique to optimize LPB. LPBSA has provided results that outperformed
popular algorithms, like the Genetic Algorithm (GA), Particle Swarm Optimization (PSO), and
even LPB. This study outlines the improved algorithm’s working procedure by providing a main
population and dividing it into Good and Bad populations, and then applying crossover and
mutation operators. When some individuals were born in the crossover stage, they have to go
through the mutation process. Between these two steps, we have applied SA using the Metropolis
Acceptance Criterion (MAC) so as to accept only the best and useful individuals to be used in the
next iteration. Finally, the outcomes demonstrate that the population is enhanced, leading to
improved efficiency and validating the performance of LPBSA.

Keywords: Algorithm Optimization, LPB, LPBSA, GA Operators

8.1 Introduction

In the realm of optimization, many algorithms display significant promise; however, they

require greater efficiency and robustness to effectively solve complex problems and obtain better

results. The LPB algorithm shows potential, while, with its exploitation and exploration, there is

room to improve or enhance its overall performance. Metaheuristics are sophisticated approaches

for addressing problems that have been created to address a variety of optimization issues. These

algorithms have demonstrated efficacy in managing intricate and insolvable scenarios, providing

an approximate resolution within a brief timeframe. In many disciplines, including engineering,

medicine, corporate planning, and energy management, optimization is essential. In addition,

metaheuristic algorithms are capable of making intelligent choices among a wide range of potential

solutions [1]. Natural systems are the source of inspiration for some popular metaheuristic

algorithms, such as the GA and PSO. The LPB algorithm was recently introduced by Rahman and

1
Rashid (2021) [2] and is intended to handle both single and multi-objective situations. Rahman

and Rashid’s study illustrates how to apply LPB in a methodical manner to maximize and acquire

optimal results. Researchers interested in creating, improving, or hybridizing the method can use

the publication as a guide as well.

The primary solution focuses on the improvement of LPB algorithm to address its limitations. To

do this we introduced LPBSA, which integrates the strengths of LPB with powerful techniques of

SA to enhance its performance. By, striking a balance between exploring new possibilities and

exploiting known ones, LPBSA provided as the best approach to the algorithm converge more

efficiently to the optimal solution. The proposed algorithm is an improvement on LPB that expands

on its capabilities. Utilizing SA as a potent method, LPBSA outperforms well-known algorithms

like GAs and PSO in terms of LPB optimization. By applying crossover and mutation operations,

splitting a main population into Good and Bad populations, and using SA with a MAC, the LPBSA

technique selects only the best candidates for the subsequent iteration. By improving the

population, this technique raises the algorithm’s efficiency and improves its performance.

Optimization algorithms are crucial in terms of solving complex or high-dimensional

problems effectively. Among those algorithms, LPB has obtained impressive results by drawing

inspiration from the academic progress of students, especially students that are currently moving

to university from high school. They are new students from college and have chosen a new field

in which they may find it difficult to improve their performance, so they need to share information

with other students in order to improve their behavior. To this end, the LPB algorithm tries to use

its methods and techniques to improve the performance of students. LPB still has room to improve

its performance, which can be achieved by integrating an additional metaheuristic technique to

address its limitation. One of the main objectives of developing LPB algorithm is to enhance

2
students' behavior with the goal of improving their overall experience across academic disciplines.

Consequently, progressions in the LPB algorithm directly contribute to the enhancement of

students' knowledge in various fields. Therefore, algorithms can be developed in order to

iteratively refine their search processes, leading to more efficient solutions. In order to strengthen

the LPB, simulating annealing was chosen suitably because it is a popular metaheuristic algorithm

that was inspired by the metallurgical annealing process. In this process, materials are gradually

cooled to obtain an optimal crystalline structure. Similarly, SA aids the algorithm escape local

optima, making it extremely effective in improving the exploration phase of the optimization

process. Also, SA increases the possibility of discovering the goal optimum because it helps in

broadening the search space and probabilistically.

8.2 Literature Review

Optimization algorithms play an important role for the complex problem solutions through a wide

range of fields. Such as, their application in education has garnered pivotal attention, as these

algorithms provide potential solutions for improving learning outcomes, support decision-making

and personalizing learning experience. Among these optimization approaches, the GA stands out

due to its capability to effectively search large solution spaces, inspired by the principles of nature

selection. GA has established to be essential in enhancing several tasks, from resources allocation

to curriculum design, highlighting its versatility and important in educational contexts. This

algorithm has also been applied to resource allocation and optimization scheduling in educational

settings. For instance, GAs have been used to mimic adaptive systems for learning, which tailors

educational content to the need of a specific learner [3]. PSO is another optimization algorithm

that was inspired by both “fish schooling” and the “social behavior of flocking birds”, and it has

been utilized for optimization tasks in numerus fields such as education. It is also absolutely useful

3
in e-learning systems, especially for some significant facilities related to these systems, including

improving the collaborative learning area and developing intelligent tutorials. One of PSO’s

aspects that makes it suitable for “real-time educational applications” is that this algorithm has a

high ability to converge fast to optimal solutions [4]. Moreover, [5] presents differential evolution

as another simple and effective optimization algorithm that has been utilized in solving many

problems in the education field, particularly for some relevant educational facilities like managing

resources, scheduling exams, and designing curriculums. Simplicity and robustness are the two

essential aspects that make this algorithm a popular choice for educational tasks.

Some approaches like “data-driven analytics”, “machine learning models”, and “education data

mining” are aiming to identify trends in student data for developing their involvement targets.

[6]states that one of the key links that will improve a student’s achievement and engagement is

personalized learning by addressing student learning styles, preferences, and wants. In addition,

by using real-time data, “adaptive learning systems” can be significantly utilized to alter the

contents and learning environment to meet the learner student’s needs. Moreover, optimization

algorithms are employed by these systems to modify feedback, assessment, and instructional

resources dynamically. Furthermore, the systems created based on these algorithms can lead to

greater student fulfillment and outcomes for improved learning [7]. Next, more of a focus will be

on existing gaps not addressed properly in previous studies.

In this section, we will highlight some significant gaps which will be focused in this study,

especially regarding personalized learning and optimization algorithms. One of the crucial gap that

has not been addressed properly is the challenge of admitting students into university programs for

which they may lack foundational knowledge. This problem highlights the need for more effective

algorithms that will assist the students to improve their skills (performance) by selecting their

4
specific learning needs. By refining these algorithms, better supports will be available for the

students in overcoming knowledge gaps and enhance their academic progress. One of the most

important areas for tracking and exposing previous algorithms is “scalability”. Many optimization

algorithms struggle during applications to large educational datasets. With existing algorithms,

there is a need to provide actionable and interpretable insights for educators. This is another gap

that should be under focus. Another gap is that, so far, there is a lack of focus on integrating

optimization algorithms for real-world educational systems and practices. LPBSA, as a novel

contribution, provides a scalable, interpretable, and integrative approach to address these gaps to

optimize learner performance. LPBSA is inspired by the process of admitting students to the

colleges or universities where they are encouraged to adapt and enhance their study skills or

behaviors. This process aligns with the essential principles of personalized and adaptive learning,

which focus on tailoring education to meet the student needs. By the incorporation of SA, which

strengthens the LPB results, LPBSA has made the algorithm more reliable and, more importantly,

it also designed to be adaptable and responsive for further use in the future, making it essential for

optimizing processes and develop outcomes in various fields like healthcare and education.

8.3. LPB

The phrases required in learner acceptance are mirrored in the LPB algorithm, which draws

inspiration from the university admissions process for recent high school graduates. These

procedures include grouping students according to their cumulative rates and subsequently

enhancing their behavior and performance after admission. These techniques also help to improve

individuals’ conduct and performance level when they are admitted to a particular department.

According to [8] [9], given the necessity for learners to adopt new study habits as they transition

from junior high to college, in order to solve this, the algorithm first chooses a subset of the

5
population. These individuals are then separated into smaller groups, from which the most

qualified applications are selected in accordance with their qualifications. Then, through

cooperative efforts that resemble cooperation and facilitate information exchange during study

sessions, their conduct and performance are further improved (referred to as crossover).

Furthermore, this approach introduces mutation, or unpredictable impacts, on their behavior. LPB

incorporates mutation and crossover strategies similar to those used in GA. Researchers used the

diagram shown in Fig 8. 1 as a flowchart to develop this algorithm.

It is important to represent how the LPB algorithm works in the next paragraph.

The initialization plays the first step in the process in which a group of people(learners) is

separated into two subgroups based on their fitness. Subsequently, work targets at enhancing

behavior through promoting working as a team and information transfer within the sub-groups.

This approach along with the development of effective study habits, is aimed at improving

individual performance and learning outcomes. Later, we have to select the best individuals from

each sub-group based on their improved behaviors.

Then we start to exchange information between selected individuals, called “crossover”, to

create new solutions. This is similar to exchanging information between parents to produce

offspring. When crossover is complete, the “mutation” operation starts, which introduces random

changes to some individuals to explore new potential solutions, and assess the fitness of the new

individuals to determine their performance. Then, the process is repeated to select the next

iteration. The process stops when a certain condition is met.

6
 Fig 8. 1: LPB Flowchart

8.4. SA

The SA algorithm was introduced in 1983 by Kirkpatrick et al. to solve the “Traveling Salesman

Problem (TSP)” [10]. SA is a probabilistic way to estimate a function’s global optimum. In

addition, it is a metaheuristic specifically designed to approximate global optimization for an

optimization problem across a broad search space, and it is an optimization technique that

maintains constant control limits while systematically modifying the variables, and upper and

lower truncation limits in a stochastic way [11].

SA algorithms draw inspiration from metallurgy’s annealing process, which involves

rapidly heating a metal followed by gradual cooling [12]. At high temperatures, atoms within the

7
metal exhibit fast movement, but as the temperature decreases, their kinetic energy diminishes.

Ultimately, this process leads to a more ordered atomic arrangement, resulting in a material that is

more ductile and easier to manipulate [12]. Numerous optimization issues, including TSP, protein

folding, graph partitioning, and job-shop scheduling, have been effectively solved with SA [13]

[14]. The capacity of SA to break free from local minima and converge to a global minimum is its

primary benefit [11]. Furthermore, SA does not require prior knowledge of the search space and

is quite simple to implement. Starting with an initial solution, the SA method iteratively improves

the existing solution by randomly perturbing it and accepting the perturbation with a predetermined

probability [15]. With more iterations, there is a progressive drop in the probability of accepting a

worse answer from its original high level [15]. Fig 8. 2 offers a through flowchart that illustrates

each phase of SA.

8
 Fig 8. 2: Simulated Annealing Algorithm Flowchart

As mentioned, SA is an optimization method inspired by the annealing process in metallurgy. It

involves heating a material and then cooling it slowly to remove defects, analogous to exploring

and refining solutions in optimization.

9
8.4.1 Key Concepts Behind SA

Some key concepts have a significant role to play in developing the techniques of the SA

algorithm. “Energy states and cost functions” is one of those concepts. In optimization, the energy

concept in a system is analogous to the cost function, that shows the value required to be minimized

or maximized. Each acceptable solution to the problem optimization corresponds to a state of the

system. And, based on cost function the quality of that solution is evaluated. Also, temperature

and cooling schedule are another two key concepts where temperatures acts as a controlling

parameter to determine the likelihood of accepting less optimal solution across the algorithms

progress, allowing the exploration of a broader solution space. Means with a higher probability of

a higher temperature allow the algorithm to accept worse solutions, promoting exploration of the

solution space. Based on a predefined cooling schedule, the temperature is regularly reduced. As

the temperature reductions, the possibility of accepting worse solution is reduced, guiding the

algorithm toward convergence at an optimal solution.

The Metropolis Acceptance Criterion is another concept and an important key to SA’s

effectiveness that allows the acceptance of worse solutions developed on the basis of a probability

that decreases with temperature assisting the algorithm in escaping local minima. LPBSA

combines the robust exploration capabilities of SA with the structured approach of LPB. So, by

incorporating the Metropolis Acceptance Criterion , LPBSA enhances its capacity to avoid local

minima and explore the solution space more effectively.

8.4.2 How the algorithm works

In the context of the algorithm’s work process, here is a step-by-step presentation of how the

algorithm works.

Initialization

10
Start the algorithm with a high temperature or cooling rate. This is known as the initial solution.

Then the cost of the initial solution will be evaluated.

Iterative Process

The following steps are repeated until the specified number of iterations is reached or the stopping

criteria are met:

✓ Generate a new solution: The existing solution is disrupted to produce a new

candidate solution and then the cost of the new solution is evaluated.

✓ Evaluate the new solution: The difference between the new solution and the current

solution in cost (ΔE) is calculated.

✓ Apply the Metropolis Acceptance Criterion: The (MAC) determines the likelihood

of accepting a solution with a higher cost, enabling the algorithm to explore a

broader solution space.

✓ Update the temperature: According to the cooling schedule, reduce the temperature.

Convergence

The proposed algorithm concentrates on refining the solution discovered during the temperature

decrease and then the algorithm becomes less inclined to accept worse solutions. When the

temperature meets the stopping condition or gets close to zero, the algorithm converges to an

optimum solution.

8.5 LPBSA

LPBSA is a hybrid optimization technique that blends SA and LPB. However, it is known as an

improvement approach that focuses more on the LPB algorithm’s optimization. By increasing

convergence, resilience, and adaptability in the solution of challenging optimization problems,

LPBSA seeks to improve on established LPB approaches. By comparing LPBSA with LPB and

11
other well-known algorithms, such as PSO and GA, LPBSA performs better through assessments

utilizing benchmark test functions. LPBSA plays a crucial role in improving the efficiency of

intelligent systems by fine-tuning parameters, lowering the complexity of computation, and

improving performance overall. This algorithm especially helpful in dynamic settings where the

optimal solution could alter over time. Because of its adaptive qualities, LPBSA can modify its

search strategies to change conditions or circumstances.

8.5.1 Practical Implementation using LPBSA to Optimize Learner

Performance

LPBSA holds essential possible for optimizing learner performance in diverse educational settings.

In this section, this algorithm’s capabilities are integrated into real-world environments.

Essentially, given the purpose of personalized learning pathways, this algorithm provides

promising chances in real-world educational settings to improve learner performance over

personalized learning pathways by analyzing individual performance data, wherever students are

struggling, and tailoring learning activities to their precise needs. The algorithm can identify such

areas, which will have a great effect in terms of improving learning experiences. There are

significant potential for the performance of optimizing learners are available in various education

settings of LPBSA, It is also has an ability to effectively applied in real world environments to

provide personalized learning pathways. LPBSA identifies areas where students are facing

challenges and tailors learning activities to address their specific requirements based on analyzing

individual performance data. As well as, this targeted approach improves learning skills and helps

in enhance overall academic outcomes. One of the most prominent implementation considerations

is “data collection”, which means it is very significant to collect comprehensive data on student

performance, such as grades, attendance, or any other relevant metrics. Optimized resource sharing

12
and adaptive learning systems, and identifying students that are currently at-risk, can be beneficial

for LPBSA, which is crucial because it will adjust the difficulty of lessons based on real-time

feedback from the algorithm, dynamically ensuring that each student is challenged appropriately.

In terms of implementation considerations, LPBSA should be integrated with learning

management systems (LMS) in order to adjust learning materials automatically.

The primary objectives of this altered algorithm are to improve convergence, adaptability and

robustness. By integrating LPB with SA the approach improves global exploration, serving to

avoid local optima. And, introduces temperature-controlled evolution that allows the acceptance

of less optimal solutions when beneficial and combines SA's exploration abilities with adaptive

learning to create an effective optimization strategy. Which is ensuring robustness throughout

several problem domain. The following section explains how LPBSA works.

LPBSA starts by initializing a population of solutions and calculating the fitness of each

individual. It then selects a subset of individuals, sorts them into Good and Bad groups based on

fitness, and assigns each individual to a group. Next, it selects individuals for crossover, ensuring

the ideal group is not empty, and performs crossover to create new individuals. Mutation is applied

to introduce random changes. The algorithm uses the Metropolis Acceptance Criterion to accept

or reject new individuals based on a random number and temperature rate. The population is

updated with the accepted individuals and the process is repeated for a specified number of

iterations or until a termination condition is met. The optimal solution is returned at the end. Fig

8. 3 shows the flowchart of LPBSA.

13
 Fig 8. 3: LPBSA Flowchart

The Metropolis Acceptance Criterion is applied before proceeding to the next iteration in LPBSA

during the mutation process. The criterion is defined in Equation 8. 1. By calculating the

probability of accepting the worse solution, this equation helps in prevent premature convergence

and maintain diversity within the population by calculating the probability of accepting less

optimal solutions.

𝐶𝑜𝑠𝑡(𝑛𝑒𝑤) − 𝐶𝑜𝑠𝑡(𝑐𝑢𝑟𝑟𝑒𝑛𝑡)
𝑃(𝑎𝑐𝑐𝑒𝑝𝑡 𝑤𝑜𝑟𝑠𝑒) 𝑒𝑥𝑝 (− ) Equation 8. 1
𝑡𝑒𝑚𝑝𝑒𝑟𝑎𝑡𝑢𝑟𝑒

8.6 Key Steps of Crossover and Mutation in LPBSA

There are some reasons behind crossover and mutation operators being used with LPBSA, such as

“exploration and exploitation”. Crossover mainly enables exploration by sharing information from

14
diverse regions in the search space. On the other hand, mutation facilitates exploitation by

implementing random and small modifications, which fine-tune solutions. Hence, these two

operators together balance exploitation and exploration, which is very significant in order to

provide effective optimization. [16] states that, by properly balancing two modes like exploitation

and exploration, increase the chances to achieve robust results. And, Diversity maintenance is

important for preserving genetic diversity within the population, by leveraging both crossover and

mutation. Which is allowing the algorithm to reduce the risk of getting trapped in local optima.

So, this diversity is crucial for achieving global optima which is the ultimate goal of the

optimization process.

Regarding “adaptability”, LPBSA can adapt to various optimization landscapes by combining

crossover and mutation. The algorithms capability to fine-tune solutions within the fitness

environment is ensured through the mutation process.

8.6.1 Crossover Operator

In LPBSA the crossover operator is inspired by the process of integrating the behaviors of

individuals to deliver a new individual with possibly superior characteristics. In LPBSA, the

crossover operator is utilized to combine information from two solutions to create a new candidate

solution, aiding to explore the search space more effectively. Two parents are selected for

crossover:

i. Parents are introduced as a decimal number that is converted to binary.

ii. The left half of the binary number for the first parent is merged with the right half of

the binary number for the second parent, and the right half of the binary number for the

first parent is integrated with the left half of the binary number for the second parent.

iii. Two new offspring are produced.

15
8.6.2 Mutation Operator

In LPBSA, the mutation operator provides a random change to the individual solution, which

means more improvement in selected individuals. This promotes variety within the population and

aids the avoidance of premature convergence. The mutation process is performed simply as

presented below.

i. New offspring from crossover are selected for mutation.

ii. A binary number is utilized for modification.

iii. One point of the binary number is randomly changed, like one 0 point changed to 1.

iv. New and improved offspring are delivered.

8.7 Case Study

Consider the following function: f(x) = x12 + x22, for integer x, 0 ≤ x1 ≤ 9000 and 0 ≤ x2 ≤ 9000.

Step 1: Create a random population (Bi) of between 0 and 9000 individuals and then provide an

average for the fitness of X12and X22, as shown in Table 8. 1.

Table 8. 1: Main population (Bi)

Bi X1 X2 X12 X22 Fitness of X12and X22

B1 4320 3120 18662400 9734400 28396800
B2 1233 4523 1520289 20457529 21977818
B3 5100 3209 26010000 10297681 36307681
B4 4355 5210 18966025 27144100 46110125
B5 2331 4266 5433561 18198756 23632317
B6 2040 2755 4161600 7590025 11751625
B7 5043 1977 25431849 3908529 29340378
B8 3460 4781 11971600 22857961 34829561
B9 1920 5510 3686400 30360100 34046500
B10 4222 3741 17825284 13995081 31820365
B11 5401 1740 29170801 3027600 32198401
B12 3351 2850 11229201 8122500 19351701

16
 B13 5201 4989 27050401 24890121 51940522
B14 2188 3477 4787344 12089529 16876873
B15 3409 1877 11621281 3523129 15144410
B16 4560 2776 20793600 7706176 28499776

Average: 28889053

Step 2: Create subpopulation (s) randomly from the main population by selecting eight individuals

and put them in descending order. Divide those eight individuals into two groups (Good) and

(Bad). Select individuals from highest to lowest from those groups, which is shown in Table 8. 2.

Table 8. 2: K Population

Bi Ki Fitness Groups

Good
B3 K1 36307681
B8 K2 34829561
B11 K3 32198401
B1 K4 28396800

Bad
B5 K5 23632317
B12 K6 19351701
B15 K7 15144410
B6 K8 11751625

Highest fitness for the Good group = 36307681

Highest fitness for the Bad group = 23632317

Step 3: Compare the main population (Bi) individuals to the highest fitness of the Good and Bad

group in Ki, which is presented in Table 8. 3.

• If the fitness values of Bi <= highest fitness of the Bad group, move Bi to the Bad group;

• else if the fitness value of Bi <= highest fitness of the Good group, move Bi to the Good

group;

17
 • else if the fitness value of Bi > highest fitness of the Good group, move Bi to the Ideal

group.

Table 8. 3: Comparison of Bi individuals with Ki highest Good and Bad fitness

Bi Fitness Fit (Bi)<= Fit (Ki5) Fit (Bi)<= Fit(Ki1) Ideal group
B1 28396800 Good
B2 21977818 Bad
B3 36307681 Good

B4 46110125 Ideal
B5 23632317 Bad
B6 11751625 Bad
B7 29340378 Good
B8 34829561 Good
B9 34046500 Good
B10 31820365 Good
B11 32198401 Good
B12 19351701 Bad

B13 51940522 Ideal

B14 16876873 Bad
B15 15144410 Bad
B16 28499776 Good
Step 4: In this step we have to select 4 individuals from the main population, ensure that the Ideal

group is not empty; if it is, choose individuals from the Good group. If the Good group is also

empty, select individuals from the Bad group. Table 8. 4 displays the selected individuals from

Table 8. 3, these four individuals are used for the next phase (Crossover stage).

Ensure that the number of selected individuals matches the specified requirement N = 4, as stated

in the initial step.

Table 8. 4: Selected Individuals

Proper individuals are selected to crossover from Table 8.3

B13 51,940,522 Ideal
B4 46,110,125 Ideal

18
 B3 36,307,681 Good
B8 34,829,561 Good

Step 5: Table 8. 5 presents the applied crossover between the individuals selected from Table 8.4

with Ki individuals from Table 8.2.

Table 8. 5: Crossover operation

IdCh NInd X1 X2 Binary (X1) Binary(X2)

E1 B13 5201 4989 1010001010001 1001101111101
E2 K1 5100 3209 1001111101100 110010001001
C1 New 5228 2441 1010001101100 100110001001
C2 New 5073 6525 1001111010001 1100101111101
E3 B4 4355 5210 1000100000011 1010001011010
E4 K2 3460 4781 110110000100 1001010101101
C3 New 2180 5165 100010000100 1010000101101
C4 New 6915 4826 1101100000011 1001011011010
E5 B3 5100 3209 1001111101100 110010001001
E6 K3 3351 2850 110100010111 101100100010
C5 New 2519 3234 100111010111 110010100010
C6 New 6764 2825 1101001101100 101100001001
E7 B8 3460 4781 110110000100 1001010101101
E8 K4 4320 3120 1000011100000 110000110000
C7 New 7008 2416 1101101100000 100101110000
C8 New 1028 6189 10000000100 1100000101101

Step 6: For the new individuals (children), apply mutation and convert one 0 to 1 for each

individual randomly in order to maximize the function. This process is shown in Table 8. 6.

Table 8. 6: Apply Mutation

IdCh X1 X2 Binary (X1) Binary(X2)

C1 5228 2441 1010001101100 100110001001
C2 5073 6525 1001111010001 1100101111101
C3 2180 5165 100010000100 1010000101101
C4 6915 4826 1101100000011 1001011011010
C5 2519 3234 100111010111 110010100010
C6 6764 2825 1101001101100 101100001001

19
 C7 7008 2416 1101101100000 100101110000
C8 1028 6189 10000000100 1100000101101

New individuals are shown in Table 8. 7.

Table 8. 7: New individuals after Mutation

IdCh X1 X2 Binary (X1) Binary(X2)

C1 7276 3465 1110001101100 110110001001
C2 7121 7549 1101111010001 1110101111101
C3 3204 7213 110010000100 1110000101101
C4 7939 6874 1111100000011 1101011011010
C5 3543 3746 110111010111 111010100010
C6 7788 3849 1111001101100 111100001001
C7 7024 3440 1101101110000 110101110000
C8 1540 7213 11000000100 1110000101101

Step 7: Before passing to the next iteration, apply the (MAC) formula shown in Equation 8. 1. In

this chapter, a random number = 0.6 and temperature rate = 100 are provided.

For each cost, compare the acceptance probability with the random number to determine

acceptance or rejection:

For C1: Acceptance probability = exp(-(7276 - 5228)/100) ≈ 0.6703. Since 0.6 < 0.6703,

C1 is accepted.

For C2: Acceptance probability = exp(-(7121 - 5073)/100) ≈ 0.6775. Since 0.6 < 0.6775,

C2 is accepted.

For C3: Acceptance probability = exp(-(3204 - 2180)/100) ≈ 0.7408. Since 0.6 < 0.7408,

C3 is accepted.

For C4: Acceptance probability = exp(-(7939 - 6915)/100) ≈ 0.6525. Since 0.6 < 0.6525,

C4 is accepted.

20
 For C5: Acceptance probability = exp(-(3543 - 2519)/100) ≈ 0.7288. Since 0.6 < 0.7288,

C5 is accepted.

For C6: Acceptance probability = exp(-(7788 - 6764)/100) ≈ 0.6742. Since 0.6 < 0.6742,

C6 is accepted.

For C7: Acceptance probability = exp(-(7024 - 7008)/100) ≈ 0.5391. Since 0.6 > 0.5391,

C7 is rejected.

For C8: Acceptance probability = exp(-(1540 - 1028)/100) ≈ 0.6428. Since 0.6 < 0.6428,

C8 is accepted.

Therefore, based on the comparison with the random number (0.6), C7 is rejected, whereas all

other costs are accepted. Table 8. 8 shows the individuals that are accepted and used for the next

iteration.

Table 8. 8: Accepted Individuals

IdCh X1 X2 Binary (X1) Binary(X2)

C1 7276 3465 1110001101100 110110001001
C2 7121 7549 1101111010001 1110101111101
C3 3204 7213 110010000100 1110000101101
C4 7939 6874 1111100000011 1101011011010
C5 3543 3746 110111010111 111010100010
C6 7788 3849 1111001101100 111100001001
C8 1540 7213 11000000100 1110000101101
Step 8: Calculate the new accepted individuals that are displayed in Table 8. 9.

Table 8. 9: Calculate accepted individuals

IdCh X1 X2 X12 X22 Fitness of X12 and X22

C1 7276 3465 52940176 12006225 64946401
C2 7121 7549 50708641 56987401 107696042
C3 3204 7213 10265616 52027369 62292985
C4 7939 6874 63027721 47251876 110279597
C5 3543 3746 12552849 14032516 26585365
C6 7788 3849 60652944 14814801 75467745

21
 C8 1540 7213 2371600 52027369 54398969

Step 9: Calculate the average for accepted individuals in Table 8.9, the individuals in Table 8.4

and (Good group) individuals in Table 8.2. The results are shown in Table 8. 10.

Table 8. 10: Provide an average

IdCh X1 X2 X12 X22 Fitness of X12 and X22

B13 5201 4989 27050401 24890121 51940522
K1 5100 3209 26010000 10297681 36307681
B4 4355 5210 18966025 27144100 46110125
K2 3460 4781 11971600 22857961 34829561
B3 5100 3209 26010000 10297681 36307681
K3 3351 2850 11229201 8122500 19351701
B8 3460 4781 11971600 22857961 34829561
K4 4320 3120 18662400 9734400 28396800
C1 7276 3465 52940176 12006225 64946401
C2 7121 7549 50708641 56987401 107696042
C3 3204 7213 10265616 52027369 62292985
C4 7939 6874 63027721 47251876 110279597
C5 3543 3746 12552849 14032516 26585365
C6 7788 3849 60652944 14814801 75467745
C8 1540 7213 2371600 52027369 54398969

Average: 52649382

Steps 2–9 are repeated until the desired number of iterations or the termination condition is

satisfied, at which point the optimal solution is provided. Let’s follow Steps 2–9 for the next

iteration.

Step 2: From the main population B, create a new subpopulation K which are illustrated in Table

8. 11.

Table 8. 11: Second K Sub-population

Bi K Fitness Groups
Good

B4 K1 46110125
B9 K2 34046500

22
 B10 K3 31820365
B1 K4 28396800

Bad
B2 K5 21977818
B12 K6 19351701
B14 K7 16876873
B6 K8 11751625

Highest fitness for the Good group = 46110125

Highest fitness for the Bad group = 21977818

Step 3: Compare the main population (Bi) individuals to the highest fitness of the Good

and Bad group in K, presented in Table 8. 12.

• If the fitness values of Bi <= highest fitness of the Bad group, move Bi to the Bad

group;

• else if the fitness value of Bi <= highest fitness of the Good group, move Bi to the

Good group;

• else if the fitness value of Bi > highest fitness of the Good group, move Bi to the

Ideal group.

Table 8. 12: Compare Bi individuals with Ki highest Good and Bad fitness

Bi Fitness Fit (Bi)<= Fit (Ki5) Fit (Bi)<= Fit(Ki1) Ideal group
B1 28396800 Good
B2 21977818 Good
B3 36307681 Good
B4 46110125 Good
B5 23632317 Good
B6 11751625 Bad
B7 29340378 Good
B8 34829561 Good
B9 34046500 Good
B10 31820365 Good

23
 B11 32198401 Good
B12 19351701 Bad
B13 51940522 Ideal
B14 16876873 Bad
B15 15144410 Bad
B16 28499776 Good

Step 4: : In this step we have to select 4 individuals from the main population, ensure that the Ideal

group is not empty; if it is, choose individuals from the Good group. If the Good group is also

empty, select individuals from the Bad group. The selected individuals are used in the next phase

(Crossover Operator). The results are shown in Table 8. 13.

Ensure that the number of the selected individuals matches the specified requirement N = 4, as

stated in the initial step.

Table 8. 13: Selected individuals

Proper individuals are selected to crossover from Table 8.13

B13 51940522 Ideal
B4 46110125 Good
B3 36307681 Good
B8 34829561 Good
Step 5: Apply crossover between the individuals selected from Table 8. 13 and Good group
individuals from Table 8. 11. The results of this operation are shown in Table 8. 14.
Table 8. 14: Crossover Operator

IdCh NInd X1 X2 Binary (X1) Binary(X2)

E1 B13 5201 4989 1010001010001 1001101111101
E2 K1 4355 5210 1000100000011 1010001011010
C1 New 5123 4954 1010000000011 1001101011010
C2 New 4433 5245 1000101010001 1010001111101
E3 B4 4355 5210 1000100000011 1010001011010
E4 K2 1920 5510 11110000000 1010110000110
C3 New 1088 5126 10001000000 1010000000110
C4 New 7683 5594 1111000000011 1010111011010
E5 B3 5100 3209 1010110000110 110010001001
E6 K3 4222 3741 1000001111110 111010011101

24
 C5 New 5630 3229 1010111111110 110010011101
C6 New 4102 3721 1000000000110 111010001001
E7 B8 3460 4781 110110000100 1001010101101
E8 K4 4320 3120 1000011100000 110000110000
C7 New 3488 2416 110110100000 100101110000
C8 New 4292 6189 1000011000100 1100000101101

Step 6: For the new individuals (children), apply mutation and convert one 0 to 1 for each

individual randomly in order to maximize the function. This process is shown in Table 8. 15.

Table 8. 15: Apply Mutation

IdCh X1 X2 Binary (X1) Binary(X2)

C1 5123 4954 1010000000011 1001101011010
C2 4433 5245 1000101010001 1010001111101
C3 1088 5126 10001000000 1010000000110
C4 7683 5594 1111000000011 1010111011010
C5 5630 3229 1010111111110 110010011101
C6 4102 3721 1000000000110 111010001001
C7 3488 2416 110110100000 100101110000
C8 4292 6189 1000011000100 1100000101101
New individuals are shown in Table 8. 16.

Table 8. 16: New individuals after Mutation

IdCh X1 X2 Binary (X1) Binary(X2)

C1 7171 7002 1110000000011 1101101011010
C2 6481 7293 1100101010001 1110001111101
C3 1600 7174 11001000000 1110000000110
C4 7939 7642 1111100000011 1110111011010
C5 7678 3741 1110111111110 111010011101
C6 6150 3977 1100000000110 111110001001
C7 4000 3440 111110100000 110101110000
C8 6340 7213 1100011000100 1110000101101
Step 7: Before passing to the next iteration, apply the MAC formula shown in Equation 8. 1. In

this chapter, we provide random number = 0.6 and temperature rate = 100. The results are the same

as below:

25
 For C1: Acceptance probability = exp(-(7171 - 5123)/100) ≈ 0.6703. Since 0.6 < 0.6703,

C1 is accepted.

For C2: Acceptance probability = exp(-(6481 - 4433)/100) ≈ 0.6703. Since 0.6 < 0.6703,

C2 is accepted.

For C3: Acceptance probability = exp(-(1600 - 1088)/100) ≈ 0.6065. Since 0.6 > 0.6065,

C3 is rejected.

For C4: Acceptance probability = exp(-(7939 - 7683)/100) ≈ 0.6311. Since 0.6 > 0.6311,

C4 is rejected.

For C5: Acceptance probability = exp(-(7678 - 5630)/100) ≈ 0.7061. Since 0.6 < 0.7061,

C5 is accepted.

For C6: Acceptance probability = exp(-(6150 - 4102)/100) ≈ 0.6703. Since 0.6 < 0.6703,

C6 is accepted.

For C7: Acceptance probability = exp(-(4000 - 3488)/100) ≈ 0.6055. Since 0.6 > 0.6055,

C7 is rejected.

For C8: Acceptance probability = exp(-(6340 - 4292)/100) ≈ 0.6703. Since 0.6 < 0.6703,

C8 is accepted.

Based on the comparison with the random number (0.6), C3, C4, and C7 are rejected, whereas all

other costs are accepted, Table 8. 17 shows the accepted individuals.

Table 8. 17: Accepted individuals

IdCh X1 X2 Binary (X1) Binary(X2)

C1 7171 7002 1110000000011 1101101011010
C2 6481 7293 1100101010001 1110001111101
C5 7678 3741 1110111111110 111010011101
C6 6150 3977 1100000000110 111110001001
C8 6340 7213 1100011000100 1110000101101

26
Step 8: Calculate the new accepted individuals after applying simulated annealing the results are

shown in Table 8. 18.

Table 8. 18: Calculate accepted individuals

IdCh X1 X2 X12 X22 Fitness of X12 and X22

C1 7171 7002 51423241 49028004 100451245
C2 6481 7293 42003361 53187849 95191210
C5 7678 3741 58951684 13995081 72946765
C6 6150 3977 37822500 15816529 53639029
C8 6340 7213 40195600 52027369 92222969
Step 9: Provide the sum, average, and max for the accepted individuals in Table 8. 18 as well as

the selected individuals from Table 8. 13. These results are displayed in Table 8. 19.

Table 8. 19: Provide sum, average, and max

Fitness of X12 and

IdCh X1 X2 X12 X22
X22
E1 5201 4989 27050401 24890121 51940522
E2 4355 5210 18966025 27144100 46110125
E3 4355 5210 18966025 27144100 46110125
E4 1920 5510 3686400 30360100 34046500
E5 5100 3209 26010000 10297681 36307681
E6 4222 3741 17825284 13995081 31820365
E7 3460 4781 11971600 22857961 34829561
E8 4320 3120 18662400 9734400 28396800
C1 7171 7002 51423241 49028004 100451245
C2 6481 7293 42003361 53187849 95191210
C5 7678 3741 58951684 13995081 72946765
C6 6150 3977 37822500 15816529 53639029
C8 6340 7213 40195600 52027369 92222969

Average: 55693299

27
Therefore, it is evident from the outcomes that the population has been enhanced, leading to

increased efficiency among individuals. Table 8. 20 presents a comparison of the average for each

iteration.

Table 8. 20: Comparision of Averages

Iteration N Average
Iteration 0 28889053
Iteration 1 52649382
Iteration 2 55693299
Fig 8. 4 shows the average results for each iteration and how they increased from iteration 0 to 2.

Fig 8. 4: Average results of three iterations

8.8 Comparative Analysis of LPBSA with Other Algorithms

As mentioned, LPBSA claims to outperform other algorithms. Therefore, it is crucial to present a

detailed comparison analysis utilizing particular benchmarks. In this section, more focus is on the

comparative analysis, the rationale behind the selected benchmarks, and the detailed outcomes that

were obtained.

8.8.1 Metrics for Comparison

28
The metrics discussed in this section have played a major role in determining the performance of

LPBSA. The average value is one of those metrics that can be utilized to represent the cost function

average value that was obtained over multiple runs, signifying the algorithm’s capability to deliver

promised solutions consistently. To reflect the algorithm’s reliability and stability, as well as

measure the cost function value variability, standard deviation is utilized as another metric.

Additionally, the next metric is the best value that represent the lowest cost function value, to

assess the algorithm's capability to find the optimal solution. The worst value is used as another

metric to obtain the highest cost function value to display the worst case of the algorithm’s

performance.

8.8.2 Benchmark Test Functions

To evaluate the performance of LPBSA, 19 benchmark test functions were utilized. These

functions are commonly used to measure the efficiency of optimization algorithms due to their

potential characteristics, such as dimensionality, separability, and multimodality. For example, test

function one, which is a sphere function, is a simple unimodal function with a single global

minimum, and test function two is a Rastrigin function used as a multimodal function with

numerous local minima, inspiring the algorithm’s exploration capability.

These test functions provide a comprehensive evaluation of the algorithm’s performance.

Different aspects of those algorithms are tested via those functions, such as their ability to converge

to the global minima, avoid the local minima, and maintain stability. Table 8. 21 shows a summary

of evaluating LPBSA with other algorithms. The table entries represent the average and standard

deviation of the cost function value over 30 runs for each entry.

29
Table 8. 21: Result of 19 benchmark test functions to compare LPBSA with other algorithms

LPBSA LPB DA PSO GA

TF
AVA STD AVA STD AVA STD AVA STD AVA STD
TF1 3.86896E- 7.25127E- 0.0018 0.00209 2.85E- 7.16E- 4.20E- 4.31E- 748.59 324.9
04 04 77 36 18 18 18 18 72 26
TF2 3.9134E- 2.67553E- 0.0052 0.00365 1.49E- 3.76E- 0.00315 0.0098 5.9713 1.533
03 03 38 25 05 05 4 11 58 10
TF3 15.573263 9.35452E 36.474 29.2241 1.29E- 2.10E- 0.00189 0.0033 1949.0 994.2
3 +00 88 55 06 06 1 11 03 73
TF4 0.1560362 3.49740E- 0.3938 0.13581 0.0009 0.0027 0.00174 0.0025 21.163 2.605
7 02 66 8 88 76 8 15 04 40
TF5 4.7676233 2.7775531 16.769 22.1925 7.6005 6.7864 63.4533 80.127 13330 85007
3 5 19 1 58 73 1 26 7.1 .6
TF6 0.0013538 1.83590E- 0.0020 0.00278 4.17E- 1.32E- 4.36E- 1.38E- 563.88 229.6
02 03 31 32 16 15 17 16 89 99
TF7 0.0029005 0.0014958 0.0049 0.00296 0.0102 0.0102 0.00597 0.0035 0.1668 0.072
20 89 75 5 93 93 3 83 72 57
TF8 - - - - -
191.56696 189.02 383.64 1.2E+ 164.4
3723.9685 3747.6 2857.5 7.10E+ 3407.2
8 06 66 12 78
93 5 8 11 5
TF9 0.0006765 0.0015 0.00184 16.018 9.4791 10.4472 7.8798 25.518 6.669
0.0007894
8 67 2 83 13 4 07 86 36
TF1 0.0116858 6.82155E- 0.0179 0.01353 0.2310 0.4870 0.28013 0.6018 9.4987 1.271
0 5 03 33 2 3 53 7 17 85 39
TF1 0.0625343 2.58492E- 0.0663 0.03097 0.1933 0.0734 0.08346 0.0350 7.7199 3.626
1 00 02 55 3 54 95 3 67 59 07
TF1 3.06720E- 5.69906E- 2.79E- 3.84E- 0.0311 0.0983 8.57E- 2.71E- 1858.5 5820.
2 05 05 05 05 01 49 11 10 02 21
TF1 2.63645E- 7.24801E- 0.0003 0.00051 0.0021 0.0046 0.00219 0.0046 68047. 87736
3 04 04 09 2 97 33 7 33 23 .7
TF1 0.9980000 4.51681E- 0.9980 1.26E- 103.74 91.243 135.40 130.09 21.32
150
4 00 16 04 11 2 64 06 91 03
TF1 0.0010323 6.76267E- 0.0023 0.00375 193.01 80.633 188.195 157.28 116.05 19.19
5 95 04 58 7 71 2 1 34 54 35

30
 TF1 - -
6.77522E- 2.46E- 458.29 165.37 263.094 187.13 383.91 36.60
6 1.0316000 1.0316
16 06 62 24 8 52 84 53
00 3
TF1 2.7054000 1.35504E- 0.3978 3.16E- 596.66 171.06 466.542 180.94 503.04 35.79
7 00 15 88 06 29 31 9 93 85 40
TF1 3.0000000 0.00000E 3.0001 0.00028 229.95 184.60 136.175 160.01 118.43 51.00
8 00 +00 42 3 15 95 9 87 8 18
TF1 -
3.16177E- 9.61E- 679.58 199.40 741.634 206.72 544.10 13.30
9 -3.862800 3.8627
15 07 8 14 1 96 18 16
8

8.9 Results and Discussion

The results in Table 8.21 show that LPBSA consistently achieved better performance than the

other algorithms, like PSO, across the majority of benchmark functions, GA, and the original LPB.

In addition, LPBSA achieved a lower standard deviation and superior average solution, and it also

showed its ability to deliver better solutions more consistently. Table 8.21 presents a whole map

for all the tests. The best tests are highlighted and the table shows the best outcomes of LPBSA.

Some test function results are illustrated in the following sentences. For example, in TF1, LPBSA

recorded an average of 3.87E-04 and standard deviation of 7.25E-4, outperforming LPB, which

achieved an average of 1.88E-03 and standard deviation of 2.09E-03. PSO and GA provided lower

results. PSO achieved 4.20E-18, but with a high standard deviation, and GA recorded an average

of 748.60.

In TF5, LPBSA obtained an average of 4.7676233 and standard deviation of 2.7775532.

By recording these results, LPBSA outperformed all the other proposed algorithms in this test

function. Moreover, LPBSA recorded the highest average and standard deviation compared to

other algorithms in TF9,TF14,TF15,TF16,TF18, and TF19. Even in TF17, LPBSA provided the

highest standard deviation of 1.35504E-15, and achieved the highest average by recording

31
2.705400000 compared to other algorithms, except LPB, which recorded an average of 1.35504E-

15. These results show the precision, consistency, robustness, and superior performance of

LPBSA, showcasing the algorithm’s enhanced capability to find the optimal solution with great

stability.

8.10 Conclusion

LPBSA offers a promising approach to optimizing solutions in various domains. By integrating

SA and LPB, LPBSA effectively balances exploration and exploitation, allowing for the discovery

of high-quality solutions. LPBSA’s ability to adapt to changing environments and its efficient

search capabilities make it a valuable tool for solving complex optimization problems. In addition

, LPBSA outperformed the LPB algorithm in terms of LPB optimization, indicating a notable

improvement over the LPB algorithm. A total of 19 benchmark test functions were utilized and

tested to compare the proposed algorithms. LPBSA beat well-known algorithms like PSO and GAs

by utilizing SA as a potent optimization strategy. This study’s systematic methodology, which

includes mutation, crossover, population division, and SA techniques, successfully increased the

population and boosted overall efficiency. Thus, these outcomes highlight LPBSA’s potential as a

useful tool for solving challenging issues in a variety of fields.

References
[1] K. Rajwar, K. Deep, and S. Das, “An exhaustive review of the metaheuristic algorithms for search and
optimization: taxonomy, applications, and open challenges,” Artif Intell Rev, vol. 56, no. 11, pp. 13187–
13257, Nov. 2023, doi: 10.1007/s10462-023-10470-y.
[2] S. Ahmed and T. A. Rashid, “Learner Performance-based Behavior Optimization Algorithm: A Functional
Case Study,” https://www.researchgate.net/publication/360906762_Learner_Performance-
based_Behavior_Optimization_Algorithm_A_Functional_Case_Study. [Online]. Available:
https://doi.org/10.21203/rs.3.rs-1688246/v2
[3] J. H. Holland, Adaptation in Natural and Artificial Systems. Mit Press, 1992.
[4] J. Kennedy, R. Eberhart, and bls gov, “Particle Swarm Optimization.”
[5] R. Storn and K. Price, “Differential Evolution-A Simple and Efficient Heuristic for Global Optimization
over Continuous Spaces,” Kluwer Academic Publishers, 1997. doi: 10.1023/A:1008202821328.
[6] J. F. Pane, E. D. Steiner, M. D. Baird, and L. S. Hamilton, “Continued Progress: Promising Evidence on
Personalized Learning,” 2015. [Online]. Available: www.rand.org/t/RR1365z2.

32
 [7] J. M. Keller, “Development and Use of the ARCS Model of Instructional Design.”
[8] S. Wang, F. Wang, Z. Zhu, J. Wang, T. Tran, and Z. Du, “Artificial intelligence in education: A systematic
literature review,” Oct. 15, 2024, Elsevier Ltd. doi: 10.1016/j.eswa.2024.124167.
[9] L. Cao, “College Students’ Metacognitive Awareness of Difficulties in Learning the Class Content Does
Not Automatically Lead to Adjustment of Study Strategies 1,” Australian Journal of Educational &
Developmental Psychology, vol. 7, pp. 31–46, 2007.
[10] S. Kirkpatrick, C. D. Gelatt, and M. P. Vecchi, “Optimization by Simulated Annealing,” 1983. [Online].
Available: www.sciencemag.org
[11] D. Delahaye, S. Chaimatanan, and M. Mongeau, “Simulated annealing: From basics to applications,” vol.
272, p. 978, 2019, doi: 10.1007/978-3-319-91086-4_1ï.
[12] C. Venkateswaran, M. Ramachandran, R. Kurinjimalar, P. Vidhya, and G. Mathivanan, “Application of
Simulated Annealing in Various Field,” Materials and its Characterization, vol. 1, no. 1, pp. 01–08, Feb.
2022, doi: 10.46632/mc/1/1/1.
[13] L. Hernández-Ramírez, J. Frausto-Solis, G. Castilla-Valdez, J. Javier González-Barbosa, D. Terán-
Villanueva, and M. L. Morales-Rodríguez, “A Hybrid Simulated Annealing for Job Shop Scheduling
Problem,” © International Journal of Combinatorial Optimization Problems and Informatics, vol. 10, no. 1,
pp. 6–15.
[14] F. P. Agostini, D. D. O. Soares-Pinto, M. A. Moret, C. Osthoff, and P. G. Pascutti, “Generalized simulated
annealing applied to protein folding studies,” J Comput Chem, vol. 27, no. 11, pp. 1142–1155, Aug. 2006,
doi: 10.1002/jcc.20428.
[15] G. Rajeswarappa and S. Vasundra, “Red Deer and Simulation Annealing Optimization Algorithm-Based
Energy Efficient Clustering Protocol for Improved Lifetime Expectancy in Wireless Sensor Networks,”
Wirel Pers Commun, vol. 121, no. 3, pp. 2029–2056, Dec. 2021, doi: 10.1007/s11277-021-08808-2.
[16] A. M. Ahmed et al., “Balancing exploration and exploitation phases in whale optimization algorithm: an
insightful and empirical analysis.”

33

View publication stats