Institutional-Repository, University of Moratuwa
Welcome to the University of Moratuwa Digital Repository, which houses postgraduate theses and dissertations, research articles presented at conferences by faculties and departments, university-published journal articles and research publications authored by academic staff. This online repository stores, preserves and distributes the University's scholarly work. This service allows University members to share their research with a larger audience.
 Research Publications |  Thesis & Dissertation |  E- Books |
 UoM Journal Publications |  UoM Conference Proceedings |  Articles published in Scimago's Q1 journals |  UoM Research Reports |  Other Articles authored by UoM staff |
Recent Submissions
item: Conference-Extended-Abstract
Intelligent tourism itinerary generation through natural language processing and hybrid recommendation systems
(Department of Computer Science and Engineering, 2025) Shouqi, S; Priyanayana, S; Thennakoon, SC; Vidaswin, V; Ninduwara, M; Athuraliya, CD
Traditional travel planning systems rely on rigid formbased interfaces with predefined dropdown menus, limiting user’s ability to express nuanced preferences. This constraint often results in generic itineraries that fail to capture individual travel styles, budgets, or interests. To address this gap, we developed TravelMate AI, a commercial itinerary builder that introduces two key innovations: natural language processing (NLP) for interpreting free-text trip descriptions and a hybrid recommendation system combining machine learning with large language models (LLMs). The novelty lies in enabling users to describe their travel preferences conversationally (e.g., “Cultural tour of Kandy with family, medium budget”) while maintaining structured outputs suitable for commercial deployment. By eliminating the need for form-based inputs, the system democratizes travel planning for non-technical users while retaining the precision of AI-driven recommendations. Furthermore, unlike existing commercial solutions, which either rely on structured inputs (TripAdvisor) or quizbased preference gathering (MindTrip), TravelMate AI processes unstructured text directly, enabling richer and more flexible itinerary customization. This adaptability allows users to input diverse and complex preferences without rigid constraints, making travel planning more personalized and user-friendly.
item: Thesis-Full-text
Design optimization of a grid scale hybrid energy storage system to maximize solar PV integratio
(2025) Ganege, HC; Chandima, DP
The global utilization of renewable energy is increasing as an environmentally sus- tainable approach to reduce greenhouse gas emissions, which are major contributors to air pollution and climate change, by decreasing reliance on oil-fired thermal power generation. Thisresearchaims tominimizeoil-fired thermal generationandmaximize solar photovoltaic (PV) generation through a grid-connected Hybrid Energy Storage System (HESS) that integrates lithium-ion Battery Storage (BS) and Pumped Hydro Storage (PHS). Solar energy is prioritized due to its technological maturity and cost- effectiveness compared to other sustainable energy sources. The HESS is employed to optimize storage performance, utilizing high efficiencyand energy management ca- pabilities of lithium-ion BS and the large-scale energy storage potential of PHS. Both technologies are well-established, offering reliability and low investment risks. A novel optimization algorithm, based on the Non-dominated Sorting Genetic Al- gorithm(NSGA-II),isdevelopedtominimizetotalexpenditures. Thisincludescapital, operational, and replacement costs associated with solar PV, lithium-ion BS, and PHS systems, as well as expenses related to solar power curtailment and supplementary oil-fired thermal energy. The multi-objective optimization problem is solved using the open-source Python framework Pymoo (multi-objective optimization in Python), and Python code developed and executed on Google Colaboratory (Colab). The algorithm determines optimal capacities for lithium-ion BS and solar PV systems at short-term intervals throughout the project planning horizon. The capacity of PHS is identified throughafeasibilitystudyconductedforaspecificregion,consideringgeologicalcon- straints. An innovative energy management strategy is proposed to optimize solar en- ergyutilizationwithintheHESS.Thisstrategyensurestheefficientutilizationofsolar energy, maintains grid stability, optimizes HESS operations within system limitations, and minimizes dependence on thermal power generation. The research includes a case study of the Sri Lankan power system, supported by economic and environmental assessments to identify the most sustainable scenarios. Sensitivity analysis is conducted to evaluate capacity variations under fluctuating in- terest rates. A roadmap for solar PV and BS capacity expansions is developed for two-year intervals throughout the planning horizon, along with the optimal timeline for integrating PHS. The final results indicate minimal variations in total cost, ranging from -14.05 % to 56.69 %, and an increase in solar PV generation by 2.19 % at the lowest interest rate and 5.13 % at highest interest rate. The solar PV capacities deter- mined from the optimization algorithm are compared with the projected capacities of solar power plants outlined in Sri Lankan generation expansion plan to evaluate their alignment with the renewable energy targets in the country. The results demostrats strong agreement with the national energy plan.
item: Conference-Extended-Abstract
SVM-Based signal detection for low-resolution quantized systems
(Department of Computer Science and Engineering, 2025) Luckshan, C; Amarasinghe, V; Gayan, S; Athuraliya, CD
The rapid evolution of wireless communication has led to the advent of next-generation networks, driven by the need for ultra-reliable low-latency communications (URLLC), and massive machine type communications (mMTC). With the future sixth-generation (6G), researchers are trying to move beyond millimeter wave (mmWave) into subTHz bands, [1] to enable the Ultra Massive MIMO (UM-MIMO) systems. Though these emerging technologies have significantly enhanced spectral efficiency and network capacity, they present challenges in power consumption due to the larger number of antennas, each equipped with dedicated analog-to-digital converters (ADCs). The power consumption of ADCs grows exponentially with resolution and linearly with sampling rate [2], resulting in unsustainable power demands. For instance, a system with 256 RF chains can consume over 250 W in ADC power alone, which is impractical for energy-constrained environments. To mitigate this challenge, low-resolution ADCs have emerged, while significantly reducing power consumption by limiting the number of quantization bits [3]. However, this low-resolution ADCs introduces nonlinear distortions, severely impacting receiver performance. Traditional signal processing techniques, which assume a linear input-output relationship, struggle to mitigate these distortions, necessitating novel approaches for effective detection and estimation. To this end, a maximum likelihood (ML) detector was derived [4] for the phase quantization in low-resolution ADCbased single-input single-output (SISO) systems, but it requires channel state information (CSI). Therefore, we are trying to apply Artificial intelligence (AI) to estimate the transmitted signal at the receiver without channel knowledge. Here, we applied support vector machine(SVM) based approach to detect the quantized signals at the receiver in a SISO system with n-bit ADC. It gives nearly optimum results when comparing to the ML detector in [4]. Also in the ML detector, an error in CSI causes a severe performance degradation in the system. Our method overcomes this issue by eliminating the need for CSI while achieving nearly optimal results. This work lays the foundation for AI-driven detection in complex systems like gigantic MIMO, where deriving an ML detector is infeasible and achieving accurate CSI through channel estimation is also impractical due to the presence of over two thousand antennas.
item: Conference-Extended-Abstract
Multimodal Search Exploration for E-Commerce
(Department of Computer Science and Engineering, 2025) Ratnalingam, G; Jalangan, M; Athuraliya, CD
The landscape of electronic commerce search has been changing drastically in a constant manner with an exponentially growing number of products, user-generated content and complex consumer behaviour patterns [2]. These aspects have made e-commerce search a challenging problem in order to provide accurate, relevant, and personalized search results. The challenges in e-commerce search are complex where the misalignment between visual and textual modalities in multimodal search systems may lead to poor search experiences, especially when users submit detailed, ambiguous or more natural queries [2]. This research addresses these issues by introducing an integrated approach that fuses text and image data within a unified space utilizing the ColPali mechanismwith late interaction for seamless multimodal alignment. The existing search systems are moving towards the conversational search or chatbots which uses the natural queries like “Black Jacket” or “Black Jacket with fleets” and through this multimodal system proposed, the specific and compelling use case of using VLM comes into the effect where a user can upload an image of a specific piece of clothing while asking a query as “Find me some similar jackets in black with fleets and a waterproof lining”. Hence the traditional systems will still struggle to interpret such queries holistically.
item: Conference-Extended-Abstract
Curriculum model for artificial intelligence in early childhood education in Sri Lankan Preschools
(Department of Computer Science and Engineering, 2025) Kodithuwakku, MK; Piumi Ishanka, UA; Athuraliya, CD
Early childhood education (ECE) receives significant advantages for learning experiences through Artificial Intelligence (AI) implementation from the first stages of education. Private preschool institutions in Sri Lanka require specialized AI curriculum because they control 71% of all preschool facilities. When preschoolers learn AI during this phase, they gain essential digital competencies that help them develop critical thinking abilities and problem-solving skills as well as creative thinking. The research presents an AI curriculum structure specifically designed for Sri Lankan preschool educational institutions. The curriculum structure presents AI concepts to students using age-relevant content and educational strategies which connect to their learning requirements.