Prof. Dr. Melih Günay danışmanlığındaki tezler
11 tez · Akdeniz University
PusOS: Sis bilişim için süpervizör modunda bir işletim sistemi yaklaşımı
The rise of connected devices in the Internet of Things era has accelerated industry digitization, prompting significant innovations in data collection, storage, and processing strategies. Last decade, we witnessed the handling of data produced by smart small devices has been handled within the scope of cloud computing, big data and high performance computing (HPC). However, the surging number of IoT devices and real-time needs necessitated new computing paradigms. And the new trend that has become more famous in this direction is Fog computing. Fog computing aims to reduce latency by bringing computing, storage, and networking closer to devices and end users to improve quality of service. Like every computing system, Fog computing also need to use a computing environment to meet requirements. The Linux operating system, which is widely used in cloud server systems, is used to present Fog computing infrastructure and service applications to date. Although this trend may seem like a practical solution in the current technology stack, it will not be sustainable in the long term in terms of system-level development processes and distribution deployment aspects. The fact that Fog computing has a middleware architecture necessitates the use of fewer resources and a compact operating system structure. Therefore, this doctoral research proposes PusOS, a novel Fog computing operating system designed to minimize latency and meet middleware requirements. Built on the open-source RISC-V architecture, PusOS leverages hardware-level virtualization by running solely in supervisor mode with a nanokernel handling data collection, storage, and processing – all within the kernel space, eliminating the need for a user space. This streamlined design, featuring modular domain-specific application stacks within the nanokernel, translates to a compact and easily deployable system. Experiments show PusOS significantly outperforms Linux in data collection and storage tasks, while virtualization tests confirm its effectiveness in fault tolerance and application migration. Furthermore, PusOS's open-source nature makes it a valuable test platform for further Fog computing research and development.
Cihaz seviyesinde enerji ayrıştırma için gözetimsiz yöntemlerin geliştirilmesi
Detailed feedback on electricity consumption encourages users to save energy, thereby reducing energy waste and consequently carbon emissions. However, measuring the individual consumption of each device connected to an internal electrical wiring system is costly and technically challenging. Today, many electricity meters have communication capabilities and can report measurements at least once per minute. By processing electricity meter data that covers a wide measurement range using data mining techniques, it may be possible to estimate the disaggregated consumption time series of individual devices and provide users with more insightful feedback. This thesis aims to develop an unsupervised, fast, and easily interpretable load disaggregation method that can also be applied to meter data with low reporting frequency. The method consists of several steps: data preprocessing, change point detection, feature extraction, clustering, and postprocessing. In a case study using electricity consumption data collected over more than three months from a single household, the consumption of certain devices was successfully disaggregated. The findings indicate that the proposed method can be applied even in resource-constrained environments and is capable of producing meaningful results.
Doğal dil işleme kullanılarak metin tabanlı açık uçlu sorular için cevapların otomatik notlandırılması
Within the scope of this thesis, a Python module that automatically grades the answers to text-based open-ended questions has been developed using various technologies developed on the Python programming language. During the development of the module, the comparison of the answers given by the students with the answer in the solution key was carried out using various machine learning and natural language processing techniques. The answers analyzed by the module were compared with the answer key after they were transferred to the vector plane with the help of pre-trained models. Within the scope of this thesis, the data of the Natural Science course was selected for this comparison.
Yapay zeka yöntemleri kullanarak toplu taşımada Antalya için optimum hat tasarımı
Public transportation is one of the most important services for cities. The growth of the cities constantly increases the passenger demand and the covered area for the public transportation. Increase in demand and coverage causes discomfort and high cost of service. Thus, the public transportation system has to be updated in most efficient and balanced way. With advancing data collecting and big data technologies, it is now possible to research more comprehensive optimization methods for public transportation planning. The methods introduced in this thesis can be helpful in various public transportation planning decisions regarding mostly bus lines such as Frequency Setting Problem. The aim of Boarding Pattern Clustering is clustering the bus lines with similar demand patterns through a regular day regardless of its popularity. Bus lines in same clusters can be considered together in various plannings. In Time Slot Clustering, various time slots that similar demands are determined and each time slot can have its own frequency setting to balance cost and comfort more dynamically. Alighting counts and bus occupancies of a route are estimated with probabilistic approach by using the data belongs to its reverse route.
Nesnelerin interneti tabanlı güneş enerjili su ısıtma sisteminin geliştirilmesi
Despite the advantages they offer, the unpredictable nature of hot water demand poses a significant impediment to optimizing their operation. There is an observable gap in existing research concerning the prediction of stochastic hot water demand and optimal operation of SWHSs. While several studies have shed light on the prediction of volumetric hot water usage and the optimization of water heating systems based on demand-side predictions, a comprehensive approach that integrates advanced learning techniques and the Internet of Things (IoT) for both prediction and optimization remains largely unexplored. This gap presents an opportunity to enhance the efficiency of SWHSs, especially in residential settings. This study aims to fill this lacuna by developing an IoT-based Solar Water Heating System and mainly on the axis of its supervising unit termed 'The Solar Water Heating System Controller Unit (SWHSCU)' that leverages IoT technology, deep neural networks (DNNs), and deep reinforcement learning (DRL) to ensure optimal hot water demand throughout the day. The proposed controller aims to not only expand the prediction of stochastic hot water demand but also integrate these predictions into the real-time operation of SWHSs, thereby enhancing system efficiency and operational reliability. The development process entails three core phases. Firstly, the design and installation of the SWHSs, during which two iterations were developed, each representing an advancement from its predecessor. Secondly, the development of the SWHSCU, incorporating IoT technology and advanced programming techniques. Lastly, the creation of a DRL algorithm framework supported by DNNs, utilizing a Long Short-Term Memory (LSTM) approach to handle time-series data effectively. The research focuses on enhancing the operation of residential solar water heating systems by optimizing hot water demand prediction through advanced AI and IoT methodologies. This advancement could catalyze significant improvements in the energy efficiency of solar water heating systems, resulting in the reduction of environmental impacts and energy costs. The research data is based on real data collected through sensors and the operating statuses of actuators connected to the controller unit of the second-generation solar water heating system, which was designed by the researcher and employed in the redesigned second-generation solar water heating system in the Konyaaltı district of Antalya, between March 15, 2023, and June 15, 2023. In addition, the data includes meteorological information obtained from the mgm.gov.tr website for the corresponding dates in the region where the solar water heating system was installed. Future work in this field will further enhance the viability and security of the developed SWHS, providing additional perspectives for cost reduction and data security. This study contributes to the literature by offering a novel approach to optimize the operation of residential SWHSs through advanced AI and IoT methodologies.
Linux tabanlı hafif ve açık kaynak kodlu büyük veri dağıtımı gerçeklemesi
The thesis covers the proper installation, configuration and optimization processes of the big data architectures for Linux operating system. The operating system that we will take as a core and customize is based on MILIS Operating System which is currently supported by Akdeniz University Scientific Research Projects Coordination Unit (BAP). Data platform and file system that is planned to be used in this project is well known and widely adapted Apache Hadoop. Apache Hadoop platform enables large volumes of data to be distributed across cluster of computer and disks with a basic programming framework. On top of Apache Hadoop, community-run data processing, security, access management, performance monitoring and other utility software is installed. In order to produce a stable Big Data processing software and distribution, a system should be put in place that includes the latest and compatible versions of the software packages to be assembled together. Development and automation of such software selection process and verification through unit tests is part of this research. In this research, a system image that contains the widely used big data modules is prepared for installation on virtual/physical disks.
Çevık ve optımum ulas ̧ım sıstemı olus ̧turmak
In today's world, increased and wide-spread population have increased the demand for public transportation. This study takes stop density, stop layout and passenger population of those stops into consideration and offers a better regulated public transportation net- work design that can satisfy the increased demand. In this study, boarding data is provided by the company that is in charge of Antalya's public transportation system. Remaining inputs are automatically taken from company's API service using .NETCore command line application and saved into a PostgreSQL database that is hosted on Azure. Google Colab, a Jupyter Notebook service, is then used as the development environment to pro- cess the data using Python language. After visualizing inputs such as bus routes, stop layout and passenger density on Google Maps and KeplerGL, with the use of DBSCAN and K-Means algorithms, data is clustered and a new way of connecting clusters is offered as a result of Uniform Cost Search. In the cost function, shortening the distance between clusters is assumed as a cost amplifier and increasing the passenger count is assumed to lower the cost. A good solution respect to cost function found with genetic algorithm from all triple, quadruple and quintet permutations of routes.
Masa tenisi robotu için makine öğrenmesi temelli top yörüngesi ve vuruş noktası tahmini yapan sistemin geliştirilmesi
This thesis study includes the method and dataset information to be used to estimate the trajectory of a ball thrown in a table tennis game. One of the most important problems for professional/semi-professional table tennis players is to find a trainer or partner who will contribute to their development. To develop skills against different playing styles, it is necessary to train with players of these styles. It is very difficult to find this opportunity in table tennis because it is less popular than other sports. To solve this problem, the development of table tennis robots for years has been the subject of research since the late 1980s and dozens of studies have been carried out in this field. Common problems for all methods; is to find the appropriate stroke point, racket trajectory, stroke rate, and angle for the table tennis robot that hits the ball well. In these methods, the main goal is to get the trajectory prediction at the right time. As the ball flight time can be under 200 ms for an attacking player, the trajectory prediction must end under 100 ms. The fact that there is no definitive solution to this problem in studies conducted with different methods reveals the necessity of the prediction algorithm to be more economical and time-sensitive. With this research, the most economical method and dataset structure that this method will work on will be designed by examining the methods in the existing studies.
Endüstriyel soğuk hava depolamada makine öğrenimi ile oda bazlı enerji tüketimi tahminleme
In this thesis, there are theoretical and practical studies required for room and/or warehouse-based energy consumption prediction studies in industrial cold storage. Thesis studies also cover a part of the Research \& Development Center project studies supported by the Republic of Turkey Ministry of Industry and Technology. In industrial cold storages cooled by central system cooling, energy measurement is generally not made on a room and/or warehouse basis. The absence of room and/or warehouse-based energy consumption data causes many operational works to be done inefficiently. By obtaining room and/or warehouse consumption, it becomes possible to carry out operational works based on data. In this case, as mentioned in detail in the thesis study, a great deal of energy savings can be made. Analysis studies of the Research \& Development Center project initiated in this context were carried out. As a result of the analysis study, it has been understood that the biggest unknown in obtaining the room and/or warehouse consumption of a facility cooled by the central system is the evaporator capacities. Within the scope of the thesis studies, machine learning technology has been researched in depth. With Feature Importance Scoring methods, the capacities of the evaporator units, which are the cooling components of the rooms and/or warehouses in the facilities, were predicted using different algorithms. As a result of the software development studies of the thesis, a software module that makes capacity estimation has been obtained. This module is designed to provide a web service with REST API technology. The position of this module in the Research \& Development Center project is also included as a diagram in the discussion and findings section of the thesis study.
Veri madenciliği teknikleri kullanılarak akademik yayınlar için performans değerlendirme sistemi
Research areas at large universities with a student body of 30K+ often include a wide range of disciplines from Social Sciences, Natural and Applied Sciences, Health Sciences, Fine Arts to Athletics. Therefore, it is challenge to evaluate and compare the research performance of individual researchers and departments within the university. In this study, a software has been developed to evaluate performance, support research and collaboration at Akdeniz University. The data for this study is obtained from the database of HR, Web of Science (Wos) and InCites. The data has been analyzed by data mining techniques and evaluated according to research areas. In this study, an original software for the evaluation of academic publication performance was implemented for the Higher Education Institutes. Research performance for an academic unit or staff may be accessible in real time. As publication performance is a good indicator of quality of research and impact, it can be used to determine the effectiveness of academicians and universities.
Java web teknolojileri kullanılarak esnek bilgisayar tabanlı değerlendirme sistemi implementasyonu
In the scope of this thesis, a web based flexible computer based assessment system was developed using Java web technologies. The system is designed to be flexible in order to eliminate the problems of existing systems. After the deployment on a web server of the developed system, thirty electronic exams were applied to the Akdeniz University Engineering Faculty students. The questions and tests added by the instructors to the system are analyzed according to the principles of item-response theory. The results of these analyses are presented the instructor and feedback are given about the questions. Within the scope of this thesis, introduction to programming course was selected as a reference course and its question bank was examined. According to the analysis results, it was determined that some questions were inappropriate for asking and some questions need to be revised. Researchers have investigated in many studies, whether the use of computer based assessment system has a negative impact on the student's exam success. Although it is proved that the exams in which multiple-choice questions are asked in the electronic environment have no effect on the student's exam success, there are no studies for open-ended questions. In the scope of this thesis, Natural Science course was selected as a reference course. In 2017-2018 fall semester, exams of this course were held in the paper environment with the open-ended questions. In 2018-2019 fall semester, exams were held in the electronic environment with the open-ended questions. According to the independent student t-test analysis performed on the average exam scores, there was no significant difference between the two groups ($p$=0,441). The findings of the study revealed that the use of the electronic exam system did not have a negative effect on student's success in the exams which consisted of open-ended questions. Electronic exams are usually held in computer laboratories of schools. This laboratories are designed to teach courses. Because of this, it is not difficult for students to look at each other's screens. Although it is not always possible to make changes in the settlement of the computers, it is possible to detect the students who are copying by examining the similarities of the student's answers after the exams. In the exams conducted using the developed system, the methods of Jaccard and Cosine similarity analysis were used to detect the students who were copying. These two methods were compared in terms of the results. It has been seen that the use of Cosine similarity analysis method is more suitable than the Jaccard similarity analysis method for copy detection. As a results of all these studies, by using open source technologies, a computer based assessment system designed in accordance with the use of educational institutions and covering all measurement and evaluation processes is developed.