Jaypee Institute of Information Technology, Noida

DEPARTMENT OF PHYSICS AND MATERIALS SCIENCE AND

ENGINEERING

Project Title: Hill Climbing Algorithm : Application to Physics

Submitted To : Dr. Vikas Malik

Enroll. No. Name of Student

9921103004 Swapnil Verma
9921103170 Aryan Singhal
9921103171 Mansi Rawat
9921103219 Aditya Sharma
9921103231 Aayushi

Course Name: Materials Science

Course Code: 16B1NPH532
Program: B. Tech. CSE
3rd Year 5th Sem

2023-2024
 ACKNOWLEDGEMENT

Before starting we would like to thank all those that made this project possible.

First and foremost we would like to place on record our deep sense of gratitude to Dr. Vikas
Malik, Associate Professor, Jaypee Institute of Information Technology, Noida for letting us
select this project and allowing us to practically apply the knowledge gained over the semester.
For providing us opportunities and helping us understand the problem statement and develop
effective solutions.

We would also like to thank our classmates, friends and family for their insightful comments and
constructive suggestions to improve the quality of this project work.

1
 TABLE OF CONTENTS

S. No. Topic Pg. No.

1. Acknowledgement 1

2 Introduction 3

3 Research papers 4-34

4 Conclusion of the Report 35

5 References 36

2
 1 . INTRODUCTION

In the vast landscape of artificial intelligence and optimization methodologies, the Hill Climbing
algorithm emerges as a stalwart, offering an intuitive and powerful heuristic search approach.
This project embarks on an exhaustive journey, seeking to dissect the intricate theoretical
foundations and unravel the practical implications of Hill Climbing algorithms. With a deliberate
focus on application, we delve into two distinct and complex arenas — the intricate problem of
solving the N-Queens puzzle in chess and the optimization challenges embedded within the
intricate fabric of physics.

2 . THE HILL CLIMBING ALGORITHM [5]

Introduction to hill climbing algorithm

A hill-climbing algorithm is a local search algorithm that moves continuously upward
(increasing) until the best solution is attained. This algorithm comes to an end when the peak is
reached.

This algorithm has a node that comprises two parts: state and value. It begins with a non-optimal
state (the hill’s base) and upgrades this state until a certain precondition is met. The heuristic
function is used as the basis for this precondition. The process of continuous improvement of the
current state of iteration can be termed as climbing. This explains why the algorithm is termed as
a hill-climbing algorithm.

A hill-climbing algorithm’s objective is to attain an optimal state that is an upgrade of the

existing state. When the current state is improved, the algorithm will perform further incremental
changes to the improved state. This process will continue until a peak solution is achieved. The
peak state cannot undergo further improvements.

3
Features of a hill climbing algorithm

A hill-climbing algorithm has four main features:

● It employs a greedy approach: This means that it moves in a direction in which the cost
function is optimized. The greedy approach enables the algorithm to establish local
maxima or minima.
● No Backtracking: A hill-climbing algorithm only works on the current state and
succeeding states (future). It does not look at the previous states.
● Feedback mechanism: The algorithm has a feedback mechanism that helps it decide on
the direction of movement (whether up or down the hill). The feedback mechanism is
enhanced through the generate-and-test technique.
● Incremental change: The algorithm improves the current solution by incremental
changes.

State-space diagram analysis

A state-space diagram provides a graphical representation of states and the optimization
function. If the objective function is the y-axis, we aim to establish the local maximum and
global maximum.

If the cost function represents this axis, we aim to establish the local minimum and global
minimum. More information about local minimum, local maximum, global minimum, and global
maximum can be found here.

The following diagram shows a simple state-space diagram. The objective function has been
shown on the y-axis, while the state-space represents the x-axis.

Hill Climbing Algorithm in AI

4
 fig1

A state-space diagram consists of various regions that can be explained as follows;

● Local maximum: A local maximum is a solution that surpasses other neighboring

solutions or states but is not the best possible solution.
● Global maximum: This is the best possible solution achieved by the algorithm.
● Current state: This is the existing or present state.
● Flat local maximum: This is a flat region where the neighboring solutions attain the same
value.
● Shoulder: This is a plateau whose edge is stretching upwards.
●
Problems with hill climbing
There are three regions in which a hill-climbing algorithm cannot attain a global maximum or the
optimal solution: local maximum, ridge, and plateau.

1.Local maximum

5
At this point, the neighboring states have lower values than the current state. The greedy
approach feature will not move the algorithm to a worse off state. This will lead to the
hill-climbing process’s termination, even though this is not the best possible solution.

This problem can be solved using momentum. This technique adds a certain proportion (m) of
the initial weight to the current one. m is a value between 0 and 1. Momentum enables the
hill-climbing algorithm to take huge steps that will make it move past the local maximum.

2.Plateau
In this region, the values attained by the neighboring states are the same. This makes it difficult
for the algorithm to choose the best direction.

This challenge can be overcome by taking a huge jump that will lead you to a non-plateau space.

Plateau

fig2

6
3.Ridge
The hill-climbing algorithm may terminate itself when it reaches a ridge. This is because the
peak of the ridge is followed by downward movement rather than upward movement.

This impediment can be solved by going in different directions at once.

Ridge in Hill Climbing

fig3

3 . APPLICATIONS

In this section we will summarize the applications of the Hill climbing algorithms in various
fields and aim to provide a brief overview of the work being done. These have been taken from
various research papers than have been provided in the references section if one wants to study in
more detail.

7
3.1 Application to ChessBoard Queen Problem [1]

Authors : 1Er. Vishal Khanna, 2Er. Abhishek Bhardwaj, 3Er. Sarvesh Chopra

Introduction

The n-Queens problem, involving the strategic placement of 'n' Queens on a chessboard without
conflicts, falls into the NP-class due to its high complexity. Traditional deterministic methods
struggle with its exponential nature (e.g., O(2^n) or O(n!)), necessitating heuristic approaches for
realistic timeframes. Recent AI strategies like simulated annealing, hill-climbing, genetic
algorithms, and others have been explored.

Addressing the inadequacies of conventional genetic algorithms (GA) in solving the n-Queens
problem, researchers propose a hybrid solution, merging Evolutionary Algorithms (EAs) and
Local Search (LS). This combination, known as a Memetic Algorithm (MA), aims to leverage
the strengths of both methods. The GA employs a straightforward coding scheme, and post
recombination, the LS is applied, utilizing problem-specific knowledge to refine solutions.

The MA tackles the n-Queens problem in two phases. Initially, it focuses on resolving hard
constraints, aiming for feasible solutions. In the second phase, the MA refines these solutions,
minimizing soft constraints through further evolution. The result is optimal solutions that meet
the requirements of the n-Queens problem. The effectiveness of MAs lies in balancing the
exploration abilities of EAs with the exploitation capabilities of local search techniques,
enhancing overall search efficiency.

Summary

N-Queens Problem (NQP)

The classic combinatorial problem is to place N-Queens on a chessboard so that no two attack
each other. This problem can be generalized as placing ‘n’ non attacking queens on an N N
chessboard. Since each queen must be on a different row and column, we can assume that queen
‘i’ is placed in ith column. All solutions to the NQP can therefore be represented as n-tuples (q1,
q2, …, q n ) that are permutations of an n-tuple (1, 2, 3, …, n). Position of a number in the tuple

8
represents queen's column position, while its value represents queen's row position (counting
from the bottom) using this representation, the solution space where two of the constraints (row
and column conflicts) are already satisfied should be searched in order to eliminate the diagonal
conflicts. Complexity of this problem is O(n!).

Fig 4 Chess board of size 8*8 with 8 queens

Optimization

Optimization is the process of identifying the best solution among a set of alternatives. Single
objective optimization employs a single criterion for identifying the best solution among a set of
alternatives.

1. Read the problem.

2. Reread the problem.

3. Draw a picture or graph if appropriate.

4. Identify the given information – What are the variables? – What are the constants? – Are there
any constraints? – Label the graph or picture.

5. What quantity needs to be maximized or minimized?

9
6. Find an appropriate equation for what needs to be maximized or minimized, and reduce it to
one variable.

7. Find the derivative and critical points for your equation.

8. Test your critical points and end points (where appropriate).

9. Reread the question and make sure you have answered what was asked.

N-Queens optimization using Memetic Algorithm

The Memetic Algorithm (MA) they have used to solve the above QMC n-Queens consists of the
Genetic Algorithm combined with a local search. Steps of Memetic Algorithm are given below:

10
Results: The below graphs show the results of applying the MA for the optimization of
N-Queens problem. In this research paper the tables show that MA gives best results in
comparison of GA in terms of space complexity as the number of iterations and number of
queen’s increases. In terms of optimal solutions HCL SMA and SALSMA gives best solutions
than GA. In comparison between HCLSMA and SALSMA on optimal solutions HCLSMA
provides best optimal solutions. Then convergence rate checked between these three techniques
HCLSMA has best convergence rate than SALSMA and GA on different number of iterations
and number of queens. Then time complexity calculated between three approaches HCLSMA
has less time complexity than SALSMA and GA for different number of iterations and number
of queens.

Note here -

1. HCLSMA: Hybrid Cultural Learning and Simulated Annealing Memetic Algorithm

2. SALSMA: Simulated Annealing Local Search Memetic Algorithm

3. GA: Genetic Algorithm

11
Table 1 Optimal solutions based on the number of Queens (n) and number of iterations=5.

Table 2 Optimal solutions based on the number of Queens (n) and number of iterations=8.

Table 3 Optimal solutions based on the number of Queens (n) and number of iterations=10.

Fig 5 Optimal solutions based on the number of Queens (n) and number of iterations=5.

12
Fig 6 Optimal solutions based on the number of Queens (n) and number of iterations=8

Fig 7 Optimal solutions based on the number of Queens (n) and number of iterations=10.

13
Table 4 Convergence rate based on the number of Queens (n) and number of iterations=5.

Table 5 Convergence rate based on the number of Queens (n) and number of iterations=8

Table 6 Convergence rate based on the number of Queens (n) and number of iterations=10.

Fig 8 Convergence rate based on the number of Queens (n) and number of iterations=5.

14
Fig 9 Convergence rate based on the number of Queens (n) and number of iterations=8.

Fig10 Convergence rate based on the number of Queens (n) and number of iterations=10.

Table 7 Time complexity based on the number of Queens (n) and number of iterations=5.

15
 Table 8 Time complexity based on the number of Queens (n) and number of iterations=8

Table 9 Time complexity based on the number of Queens (n) and number of iterations=10.

Conclusion of the paper

The N-Queens Problem (NQP), akin to the well-known Travelling Salesman Problem (TSP),
poses a significant challenge as an NP-hard problem. Conventional Genetic Algorithm (GA)
implementations often fall short in providing satisfactory solutions. To address this, we propose a
hybrid approach integrating an Evolutionary Algorithm for generating feasible solutions and a
Local Search for optimizing these intermediate solutions. This Memetic Algorithm (MA),
combining GA with local search, proves effective in tackling the complexity of the NQP.

The two-phase solution process of HCLSMA involves initially searching for Queens' shift
patterns that adhere to hard constraints. Notably, HCLSMA outperforms GA in terms of optimal
solutions, time efficiency, and convergence rate. Our future work aims to explore parameters
such as conflict minimization and revolution rate, facilitating a comprehensive comparison
between HCLSMA, GA, and SALSMA.

3.2 Application of iterative hill climbing to the sound speed profile

inversion in underwater acoustics [2]

16
Authors: O Zaikin and P Petrov

Introduction
This study addresses optimization challenges in fields like electrical engineering and molecular
modeling, where complex cost functions hinder traditional algorithms. It applies iterated local
search, specifically hill climbing, to acoustical parameter estimation in shallow-sea geoacoustic
waveguides. The paper demonstrates the algorithm's effectiveness in solving inversion problems,
emphasizing its practicality. Sections cover dispersion-based geoacoustic inversion, details of the
iterative hill climbing algorithm, its application to the inversion problem, and supporting
computational examples. Overall, results suggest that iterated local search successfully handles
acoustical parameter estimation in shallow-sea geoacoustic waveguides.

Geoacoustic inversion based on the dispersion data matching

The term geoacoustic inversion in the field of underwater acoustics designates a specific class of
problems related to the reconstruction of the media properties using some data on the sound
propagation through these media. A typical geoacoustic waveguide in a shallow sea is comprised
by a water column and several layers of bottom sediments where the acoustical parameters, the
sound speed c=c(z) and the density ρ= ρ(z), vary with the depth z. By contrast to the direct
problems where the media parameters are known and the acoustic pressure field has to be
computed, in the inverse problems some acoustic data is known, while the media parameters are
inverted for. Recently it was found that dispersion data could be successfully used as the input
for the inversion of the environment parameters. An important advantage of this technique is that
the dispersion data can be obtained from a single-hydrophone measurement. By contrast, other
inversion methods usually rely on the measurements performed using arrays of receivers that are
very expensive and complicated in deployment. By the dispersion data we mean hereafter the
dispersion curves (DCs), i.e. the functions ττ mf ),(= , where f is the sound frequency and m is
the mode number. In shallow-water waveguides acoustic modes with different numbers m and at
different frequencies f arrive to the receiver (which is located sufficiently far from the source) at
different moments τ. The delays of one mode with respect to another depend on the properties of
the media, i.e. the sound speeds and densities in the water column and the bottom [9]. Measuring
such delays and using the DCs as the input for the inversion algorithms we may therefore recover

17
the media parameters. Usually an inversion algorithm is based on the matching of the DCs
obtained from the experimental data and DCs computed using a certain model of the waveguide.
The experimental DCs (EDCs) are obtained from a recording of a pulse signal by a single
hydrophone (the receiver) which is located several kilometers away from the source that emits
the signal. The extraction of the DCs proposed in is based on the time-frequency analysis of the
received signal and the use of the so called warping operators that greatly enhance the possibility
of the modes separation. The theoretical DCs (TDCs) are obtained by the formula

where R is the source-receiver range, gv is the modal group velocity [9], 0 τ is the time offset
correction. The waveguide is described by the set of parameters A , and from these parameters
one can compute the group velocities(v by solving acoustical spectral problem [8, 9, 11].
Provided that EDCs τ(F,M) and TDC’s τ1(F,M) are obtained for certain vector of frequencies
f1,f2,f3,,,fn , we can compute the root mean square mismatch E(A,τ0) of the experimental and
theoretical dispersion data using by formula

where M is the number of modes detected on the spectrogram of the received signal. We can
therefore estimate the waveguide parameters A and the scalar 0 τ by the minimization of error
function E(A,τ0)

where the values * A and * τ0 constitutes the global minimum point for the function ),( 0 AE τ .
Note that parameter space in this problem can have a large number of dimensions while the cost
function usually has many local minima that deny the use of the simple local search algorithms.

18
In the papers a brute-force global search was used, in the genetic algorithm was employed, and in
the recent work a trans-D Bayesian inversion method was developed. In the present study we
restricted our attention to the discrete (finite) search space with linearly spaced trial values for
each parameter in vector A and for 0 τ .

Computational experiments
In previous work (see [14]) we developed a simple normal mode code capable of computing the modal
group velocities Afv ),( g from a given waveguide parameters A . This routine was used as a part of
brute-force search algorithm SSPEMDD [16] for the minimization of the error function E implemented
using the Message Passing Interface (MPI). For the numerical experiments described below the iterative
hill climbing algorithm suggested in the previous section was implemented using C++ and added to
SSPEMDD toolbox. It employs the same normal mode evaluation routine based on the finite differences.

Test case 1: Inversion of the sound speed profile in the water

In the first series of experiments the authors used the same synthetic dispersion data as in [14]. We
estimated the sound speed profile (SSP) in the water using the dispersion curves in the frequency range
from 150 hz to 250 hx at R=3500 We used five nodes at fixed depths z0,z1,z2,z3,z4 for the
approximation of SSP by a vector c ( ii =)( czc is the sound speed c at depth i z ) with the fixed (known)
value of the sound speed 0 c at the sea surface z0 0= . The search space for this case is presented in table
1.

Note that in this test we set 0 τ to zero and estimate the correction to the source-receiver distance R
instead. Also note that due to the processing errors the DCs extracted from a pulse signal do not match
their theoretical counterparts exactly. More details on the DCs extraction technique can be found in [7]
and references therein.
In [14] the brute-force search for this test case was carried out on 15 nodes of the “Academician V.M.
Matrosov” computing cluster (Irkutsk Supercomputing Center of SB RAS). Each computing node of this
cluster consists of two 16-core CPU AMD Opteron 6276, and therefore 480 CPU cores in total were used.

19
All 7973721 points of the search space (see table 1) were processed in 8 hours and 35 minutes, as a result
the global minimum point with E = 0.0069710 s was found. We performed 20 separate launches of the
iterative hill climbing algorithm from different randomly chosen start points, including 10 launches with
100 iterations and 10 launches with 1000 iterations. Each of these runs was executed on a single core of
CPU AMD Opteron 6276. The characteristics of the obtained 20 local minima are shown in table 2 and in
figure 1. One launch took 44 minutes on average in the case of 100 iterations and 8 hours 2 minutes on
average in the case of 1000 iterations.

Test Case 2: Bottom parameters inversion

20
In this example we estimated the bottom parameters in shallow sea (sound speed c and density ρb )
together with the arrivals offset τ0 (see equation (1)) using the dispersion data extracted from a simulated
pulse signal. We used the waveform from [17] for the emitted signal. The DCs were extracted in a
frequency range from 20 Hz to 200 Hz. The sound speed in the water column was considered constant
cw=1500m/s , and the bottom depth was set to h=90m . The forward propagation problem was solved
using our normal mode code [16] with cb=1800m/s, ρb=1.8g/cm^3 , and the signal in the receiver located
at R=8000m from the source was recovered using the standard Fourier synthesis technique [9]. The
computed signal is shown in figure 2, while the DCs are plotted in figure 3 against the spectrogram of the
computed signal (see [7, 11] for the detailed explanation of the extraction technique).

21
The search space for this example is described in table 3.

For this case SSPEMDD was launched on 10 computing nodes of the computing cluster, and therefore
320 CPU cores in total were used. All 48373 points of the search space (see Table 3) were processed in 52
minutes, and the global minimum point with E = 0.00267443 s was found. We also performed 20 separate
launches of the iterative hill climbing algorithm with different randomly chosen start points: 10 launches
with 10 iterations and 10 launches with 100 iterations. The Error function values for the obtained 20 local
minima are shown in Table 4 and in figure 4. One launch took 48 minutes on average in the case of 10
iterations and 7 hours 48 minutes on average in the case of 100 iterations (again one CPU core was used
for each launch

22
It turned out that iterative hill climbing requires significantly less computational resources (in comparison
with brute-force) to achieve a local minimum with the acceptable characteristics. In the test case 1 one
launch of iterative hill climbing with 100 iterations took on average 5597 times less resources, and 513
times less resources for one launch with 1000 iterations. In the test case 2 one launch with 10 iterations
took on average 346 times less resources, and 36 times less resources for one launch with 100 iterations.

Conclusion

This study applies the iterative hill climbing algorithm to geoacoustic inversion, demonstrating
its efficiency in quickly estimating sound speed profile (SSP) or bottom parameters. The
numerical experiments highlight its faster performance compared to an exhaustive search
algorithm, even in relatively low-dimensional spaces. A small number of iterations achieves
highly accurate approximations for waveguide parameters A. Future work involves extending the
algorithm to larger search spaces (up to 20-30 dimensions) for full-scale inversion problems,
comparing its performance against other methods in terms of accuracy and efficiency.
Additionally, enhancements to the hill climbing algorithm will include integrating metaheuristics
like simulated annealing and tabu search.

23
3.3 Enhancement of solar photovoltaic using maximum power point tracking
based on hill climbing optimization algorithm [3]

Authors : A. Ulinuha and A. Zulfikri

Introduction

Electrical power is expected to be the main and important commodity in the future. Some
reasons behind this prediction are the ease of generation using various resources, the ease of
transmission, the extensive use for different purposes and the increasing price of fossil fuels is
another reason why people move to use electricity. As a result, the demand for electricity tends to
increase in the future. With a rising demand for electricity and the escalating cost of fossil fuels,
there is a shift towards renewable resources for electricity generation. Renewable energy,
particularly solar photovoltaic, is recognized as a key solution for sustainable energy supply. The
advantages of solar photovoltaic include simplicity of energy conversion and flexibility in
construction.

The voltage generated by the panel depends on sunlight intensity and, for the purpose of
preserving the generated voltage, the intensity of the light must be maintained high. To maintain
the voltage, the panel must be directed so that it stays facing toward the sun. The scheme of
Maximum Power Point Tracking (MPPT) is applied to direct the panel for maximum capture of
sunlight. This study proposes the method of Hill Climbing to search the most optimal panel
position that enables providing maximum voltage.

Summary

Maximum Power Point Tracking

The electrical behavior of a Photovoltaic (PV) array, when subjected to the same irradiance,
exhibits a current-voltage characteristic, revealing a distinct point known as the maximum power
point (MPP). To enhance power output, Maximum Power Point Tracking (MPPT) techniques are
employed, continuously monitoring and adjusting the system to operate at the optimal VMPP
(voltage) or IMPP (current) that yields the maximum power output, PMPP, considering real light

24
intensity and ambient temperature. It's important to acknowledge that under partial shading
conditions, multiple local maxima may arise, but there is ultimately only one true MPP. Figure 1
illustrates the typical relationship between Power (P), Voltage (V), and Current (I)

Fig 11

To identify the Maximum Power Point (MPP), it is necessary to simultaneously consider Voltage
and Current. Typically, sensors for voltage and current are utilized, and power is subsequently
computed from these measurements. Various Maximum Power Point Tracking (MPPT) methods
have been developed, each differing in complexity, required sensors, convergence speed, cost,
effectiveness range, hardware implementation, popularity, and other factors. In this paper, the
Hill Climbing method is employed to guide the array towards the MPP. The choice of this
method is dictated by the nature of the problem, assuming its suitability for determining the MPP
in this context.

Hill Climbing

Hill climbing is an optimization algorithm designed to construct a search trajectory in the search
space, aiming to achieve the optimal condition. The process commences with an arbitrary
solution, and incremental changes are made to iteratively search for a better solution. This search
is repeated, and upon detecting a more optimal condition, the solution is updated accordingly.
The algorithm, known for its simplicity, is effective in finding optimal solutions for convex
problems.

25
From a mathematical perspective, Hill climbing seeks to maximize (or minimize) a target
function f(x), where x represents a vector of continuous and/or discrete values. In each iteration,
Hill climbing adjusts a single element in x and assesses whether this change enhances the value
of f(x). The algorithm accepts any change that improves f(x), and this process continues until no
further changes can be made to enhance the value of f(x). Applied to the current problem, Hill
Climbing is utilized to ascertain the optimal position of a photovoltaic panel, ensuring it
maximizes power output.

Hill Climbing is chosen for its suitability in solving the convex problem of finding the Maximum
Power Point (MPP), which typically has a single optimal position. The system employs a
closed-loop control mechanism based on the position that yields the maximum power. The
process involves updating the position if the power at the new location surpasses the current one,

26
iteratively continuing until no further improvement in power is possible. Once the MPP is
achieved, the panel halts.

The system is designed to initiate panel movement every 30 minutes, considered a reasonable
interval for potential discovery of a new MPP. This periodic movement also serves to minimize
power consumption by the motor and controller. The flowchart depicting the Maximum Power
Point Tracking (MPPT) process for Solar Photovoltaic using the Hill Climbing method is
illustrated in Figure 2.

To implement the Maximum Power Point Tracking (MPPT) using the Hill Climbing method, a
50 WP solar photovoltaic system serves as the plant. Energy conservation is achieved through
the use of a 12V/32Ah battery, charged via a 12V/20A solar charger controller. Voltage and
current sensors are employed to measure the respective values, allowing the processor to
calculate power. An Arduino Microcontroller processes the data and issues commands to the
motor for rotation. The DC motor PG45, driven by the L298N motor driver, facilitates the
adjustment of the array to attain the Maximum Power Point (MPP). The developed system
underwent real-world testing conducted over several days, operating from 08:00 a.m. to 16:00
p.m. The experiment compared the system's performance under a fixed position against one
driven by the Maximum Power Point Tracking (MPPT) using the Hill Climbing algorithm, as
outlined in Table 1. Generally, the Hill Climbing algorithm demonstrated the ability to generate
higher power. Figures 5, 6, and 7 display the voltage (V), current (I), and power (P) respectively,

27
affirming that the PV array equipped with MPPT exhibited superior performance. However,
during certain periods, particularly in the morning, the method faced challenges in generating
higher power, attributed to difficulties in distinguishing optimal positions due to slight variations
in light intensity. This supports the earlier explanation that under partial shading conditions,
MPPT may encounter issues with multiple optimal positions, potentially leading to the algorithm
being trapped in a local maximum. Despite these challenges during specific time periods, Hill
Climbing consistently outperformed the fixed array position in terms of voltage, current, and
power throughout the optimization period.

28
During the operational observation of the photovoltaic (PV) panel utilizing Maximum Power
Point Tracking (MPPT), the panel continuously adjusted its position, with power increments
visually displayed on a dot matrix. The panel typically stopped at the position providing
maximum power, taking 10 to 15 seconds to reach the Maximum Power Point (MPP). The PV
system's performance summary is as follows:

Incorporating the energy consumption by the controller and actuator, which draw currents of 0.2
Amp and 0.42 Amp, respectively, at 12 Volts, with each movement taking 15 seconds for a total
of 17 movements, results in an energy consumption of 0.527 Joule. Despite this, the system still
provides a net energy of 14.033 Joule, surpassing the energy without MPPT, which is 12.99
Joule.

29
For the developed system, controller and actuator are employed, and the required power and
energy must be calculated. The controller and actuator take current of 0.2 Amp and 0.42 Amp,
respectively for the voltage of 12 Volt. The time required for every movement is 15 seconds and
for the total switching of 17 times. Based on these data the energy drawn by MPPT is:

Ec&a = 17 x (0.2 + 0.42) x 12 x 15/(60 x 60) = 0.527 Joule

Where Ec&a is Energy consumed by controller and actuator. It may be concluded that taking into
account the energy drawn by the controller and actuator enables the system still providing higher
energy with the net energy of 14.56 – 0.527 = 14.033 Joule, higher than that without MPPT of
12.99 Joule.

Conclusion of the paper

Solar Photovoltaic (PV) systems are widely employed for electricity generation, with power
output fluctuating based on changing light intensity. To maintain consistent power generation,
the solar panel's position is adjusted to face the sun, facilitated by Maximum Power Point
Tracking (MPPT). In this system, a microcontroller calculates feedback from voltage and
current, determining the optimal position for maximum power. The Hill Climbing algorithm is
utilized for this purpose. Despite the energy consumed by the microcontroller and actuator,
measurements and calculations confirm that the system equipped with MPPT provides higher
energy compared to a fixed system.

3.4 Hill-Climbing for a Noisy Potential Field Using Information Entropy [4]

Authors: Piljae Kim , Satoru Nakamura and Daisuke Kurabayashi

Introduction

For a robot navigation system used in an unpredictable environment, it is generally effective to create a
pathway that robots can track for carrying out a given task, such as reaching a goal. In the biological
world, ants construct a foraging path using a volatile substance called a pheromone, which has been
widely studied and whose characteristics have been used in a transportation network model. When a
navigation path is constructed by autonomous agents using this pheromone model, the created potential

30
field can be very noisy, with many local peaks due to the unsynchronized updates of the field. In this
paper, a new hill-climbing algorithm is proposed. The algorithm minimizes information entropy and can
track dynamic and noisy potential fields. The proposed algorithm is evaluated through a computer
simulation.

Summary

In recent years, there has been a growing interest in leveraging robotic technology, particularly in
hazardous and disaster-affected areas. However, challenges arise when robots navigate through
uncertain and dynamic environments, hindering their ability to perform tasks effectively. This
paper addresses the difficulty robots face in tasks such as climbing potential fields due to the
high uncertainty of their position and the unpredictable nature of the environment.Inspired by the
efficient foraging behavior of ants using pheromones to establish local pathways, the authors
propose a novel algorithm for climbing noisy potential fields. The algorithm is based on the
concept of infotaxis, originally developed to solve the problem of searching for odor sources
with limited information. Unlike traditional gradient-based methods, infotaxis counts the number
of encounters with an odor plume, allowing the agent to navigate without relying on gradient
information.The authors adopt a transportation network model with RFID tags as an information
medium, as previously proposed by Herianto et al. However, the constructed potential field in
their study presented challenges, including multiple local peaks, making traditional hill-climbing
methods impractical. To address this, the paper introduces a new climbing algorithm designed to
navigate the noisy potential field effectively.The algorithm utilizes Shannon's information
entropy to calculate the uncertainty of the pheromone source's location based on locally acquired
information. By minimizing entropy during the climbing process, the algorithm probabilistically
narrows down the position of the source. The proposed method is evaluated through computer
simulations involving reaching the pheromone source in a given potential field. While initially
assuming unimodality in the algorithm's development, it proves effective for various potential
field scenarios, including In this study, we consider an independent agent that climbs the
pheromone potential field using local input information. The poten-tial field can be traced by the
agent for determining the location of the Figure 4.Flow diagram of the action-decision process
pheromone source and be used for navigation inside the field. Since the actual robot is used as an
agent, various restrictions may exist when the agent searches for a peak in the potential field,

31
which is diɼerent from the traditional solution-searching algorithms. For example, a robot can
only observe the current pheromone value from a given position, thus itis impossible to use
derivative values in the same way used in the steep-est descent method. Furthermore, the
sampling cost becomes larger when compared with the other solution-searching algorithms,
because it is necessary to gather density information by actually moving around the field when
the agent is in searching mode.In this study, the following prerequisites are given to the
agent.·Only local information is observable to the agent.·There is no sharing of information
among agents.In other words, only partial information inside the observable range,which is very
small in comparison with the size of the work space, is ac-quired by the agent. We assume that
the information on the pheromone density is only read from a RFID tag within the observable
range.As for the potential field, the observation points, each of which stores a scalar value of the
potential field, are to be arranged on the grid. A uni-modal potential field that has a goal at the
center is also assumed. The Observation points change the density value through the interactions
with the agent.3.2. The Action Decision AlgorithmAs we mentioned earlier, for a dynamic and
noisy potential field, it is not possible to use the method of tracing the gradient from the observed
scalar values. This is caused by the fact that local peaks are scattered over the field, which means
the agent falls into local minima or maxima.What is more, even in the event that the agent
manages to escape from those local solutions by a random action, it is not guaranteed that the
agent will keep tracing the slope until it reaches the global peak, simply because the random
movement makes the agent highly likely to follow a diɼerent direction to the global
maximum.This problem, therefore, requires an algorithm that is able to stochasti-cally estimate
the position of the peak until the agent has reached the destination. We adopt Bayes’ theorem for
this task so that each agent maintains the existence probability of the source as a probability
den-sity distribution, which corresponds to the mapping part of the action decision process in
Fig.4

32
.On the other hand, for the agent’s action planning algorithm, the con-cept of information
entropy has been adopted. Qualitatively speak-ing, information entropy is large when the
probability density distribu-tion spreads uniformly, i.e., having the same probability as those with
multi-pheromone sources or different distributions.

Conclusion of the paper

In this paper, a new hill-climbing method, which makes use of infor-mation entropy, is proposed
for dealing with a noisy pheromone po-tential field in order to build a guidance system for a
robot. Overall,the proposed algorithm showed improved climbing performance for the
pheromone potential field when compared with other random-based methods. Interestingly, the
designed algorithm also provides an action pattern similar to the searching behavior of insects
studied in the past.In spite of the satisfactory results, the algorithm still requires some prior
information such as a likelihood function. A number of studies on chem-ical plume tracing
assume an approximate physical model for the po-tential field and have been presenting
successful tracing results. This Can be explained from the fact that the probabilistic methods are
com-paratively robust for modeling errors or external noise. However, as we presented in this
paper, an estimate of the probability distribution of the potential field is also possible, provided
that the sampling data are ad-equately prepared. We are now further considering how to
dynamically adjust the required likelihood function while carrying out the climbing tasks.
Through this kind of a dynamic learning procedure, the agent will also be able to climb a fully
unknown potential field while observing the environment.In addition, as a unique yet heuristic

33
design scheme in this study, some parameters such as Offsetor the weight function have not been
fully evaluated, and remain to be examined in further work.Finally, the proposed hill-climbing
algorithm is also applicable to other general potential fields. This has been confirmed in our
simulation, although it was not presented in this paper.

34
 CONCLUSION

In conclusion, the application of the Hill Climbing algorithm to the chess board-queen problem
and its extension to diverse fields, such as physics, demonstrates its versatility and effectiveness
in solving complex optimization challenges.The exploration of the N-Queens problem
highlighted the algorithm's capability to navigate through the solution space efficiently. The
chessboard-queen problem, a classic example of combinatorial optimization, showcased how
Hill Climbing can be employed to find a satisfactory arrangement of queens on the chessboard,
avoiding mutual attacks. This application not only illustrated the algorithm's adaptability to
discrete problem spaces but also emphasized its relevance in addressing real-world problems
with intricate constraints.Furthermore, the extension of Hill Climbing to the realm of physics, as
exemplified in research papers like "Enhancement of Solar Photovoltaic Using Maximum Power
Point Tracking Based on Hill Climbing Optimization Algorithm," signifies the algorithm's
potential in optimizing continuous and dynamic systems. In this context, the algorithm was
employed to enhance the performance of solar photovoltaic systems by optimizing the maximum
power point tracking. The utilization of Hill Climbing in this scenario showcased its ability to
adapt to noisy potential fields and navigate through complex and fluctuating environments,
thereby contributing to the advancement of renewable energy technologies.The research paper
titled "Hill-Climbing for a Noisy Potential Field Using Information Entropy" further underscored
the algorithm's robustness in scenarios with noisy data. By incorporating information entropy, the
algorithm demonstrated resilience in navigating through uncertain and dynamic landscapes. This
adaptability is particularly valuable in real-world applications where external factors may
introduce variability and unpredictability.In summary, the Hill Climbing algorithm, applied to the
chess board-queen problem and extended to physics-related challenges, showcases its efficacy in
solving optimization problems across diverse domains. The algorithm's ability to handle discrete
and continuous problem spaces, navigate noisy environments, and adapt to dynamic conditions
positions it as a valuable tool for addressing complex real-world problems in various fields. As
we continue to explore and refine optimization algorithms, Hill Climbing stands out as a reliable
and versatile approach with broad applicability.

35
 REFERENCES

[1] E. Vishal Khanna, E. A. Bhardwaj, E. S. Chopra, and A. Prof, “N-Queens problem

optimization using various memetic algorithms,” Ijedr.org. [Online]. Available:
https://www.ijedr.org/papers/IJEDR1702196.pdf. [Accessed: 29-Nov-2023].

[2] O. Zaikin and P. Petrov, “Application of iterative hill climbing to the sound speed profile
inversion in underwater acoustics,” IOP Conf. Ser. Mater. Sci. Eng., vol. 173, p. 012022, 2017.

[3] A. Ulinuha and A. Zulfikri, “Enhancement of solar photovoltaic using maximum power
point tracking based on hill climbing optimization algorithm,” J. Phys. Conf. Ser., vol. 1517, no.
1, p. 012096, 2020.

[4] P. Kim, S. Nakamura, and D. Kurabayashi, “Hill-climbing for a noisy potential field
using information entropy,” Paladyn, vol. 2, no. 2, 2011.

[5] O. Mbaabu, “Understanding hill climbing algorithm in artificial intelligence,”

Engineering Education (EngEd) Program | Section. [Online]. Available:
https://www.section.io/engineering-education/understanding-hill-climbing-in-ai/. [Accessed:
29-Nov-2023].

36