Development and research of an intelligent system for predicting natural phenomena
Students Name: Nahachivska Solomiia-Yuliia Yaroslavivna
Qualification Level: magister
Speciality: Information Technology Design
Institute: Institute of Computer Science and Information Technologies
Mode of Study: full
Academic Year: 2025-2026 н.р.
Language of Defence: ukrainian
Abstract: Nahachivska S.-Y. Ya., Oborska O.V. (Supervisor). Development and Research of an Intelligent System for Predicting Natural Phenomena. Master’s Qualification Thesis. – Lviv Polytechnic National University, Lviv, 2025. In the context of increasing climatic instability and the unpredictability of natural disasters (floods, storms, droughts), ensuring highly accurate and timely forecasting is critically important. The effectiveness of preventive measures and the minimization of socio-economic and human losses directly depend on the ability to operatively process and integrate large arrays of heterogeneous data (meteorological parameters, satellite images, IoT sensor data). Traditional physical and mathematical models are often computationally expensive and slow, which limits their efficiency in scenarios requiring rapid response. Object of research: Processes of multi-factor data collection, pre-processing, and intelligent analysis aimed at increasing the reliability and speed of natural phenomena prediction. Subject of research: Software and architectural design and implementation of an intelligent system utilizing deep machine learning methods for forecasting and supporting decision-making related to emergency prevention. Goal of research: To create a scalable information system with an optimized time series processing model that applies Deep Learning algorithms to identify complex hidden patterns and generate accurate short- and medium-term prognostic estimates. Novelty: A hybrid neural network architecture has been developed and experimentally validated, optimized for analyzing spatial data (including imagery), which increases the accuracy of forecasting natural disasters compared to classical models. A heterogeneous data fusion method has been proposed, utilizing adaptive weighting of embedded layers to unify data from IoT sensors, meteorological stations, and satellite imagery, enabling a comprehensive risk assessment. The main outcome is the creation of a functional prototype of an intelligent system that successfully applies deep learning principles to model nonlinear dependencies in large-scale time series. It has been demonstrated that the developed system outperforms traditional forecasting models in terms of both accuracy metrics and prediction generation speed. The practical value lies in reducing the time required for analytical data processing and providing emergency services with up-to-date, scientifically grounded information for prioritizing regions and protecting lives and property. Keywords: Intelligent system, forecasting, natural phenomena, machine learning, deep learning, neural networks, Big Data, time series, Python, LSTM. List of used literature sources: Krizhevsky A., Sutskever I., Hinton G. ImageNet Classification with Deep Convolutional Neural Networks. NIPS, 2022. Goodfellow I., Bengio Y., Courville A. Deep Learning. MIT Press, 2025.