02. Master's Thesis

Permanent URI for this collection

https://nur.nu.edu.kz/handle/123456789/8136

Browse

Browsing 02. Master's Thesis by Author "Abdek, Meiirkhan"

Now showing 1 - 1 of 1
  • No Thumbnail Available
    ItemOpen Access
    AI BASED TECHNIQUES FOR SHORT-TERM LOAD FORECASTING
    (Nazarbayev University School of Engineering and Digital Sciences, 2025-04-29) Abdek, Meiirkhan
    This thesis investigates the application of artificial intelligence-based techniques for short-term load forecasting (STLF) using synthesized data representing Kazakhstan's electrical grid. The research compares traditional statistical approaches, machine learning algorithms, and deep learning methods to determine optimal forecasting solutions for the region's specific load patterns. Five models were implemented and evaluated: Seasonal Autoregressive Integrated Moving Average (SARIMA), gradient boosting methods (LightGBM and XGBoost), and recurrent neural network architectures (Long Short-Term Memory and Gated Recurrent Unit). Following extensive exploratory data analysis and feature engineering, the study reveals that gradient boosting methods significantly outperform both statistical and deep learning approaches. XGBoost achieved the best performance with a Mean Absolute Percentage Error (MAPE) of 7.15%, closely followed by LightGBM at 7.42%. The deep learning models performed moderately well (LSTM: 11.73% MAPE, GRU: 12.21% MAPE), while SARIMA showed considerably poorer results (20.85% MAPE). Analysis of temporal patterns revealed strong hourly dependencies but relatively weak weekly and monthly seasonality in Kazakhstan's load profiles. This characteristic partly explains the superior performance of gradient boosting methods, which excel at capturing complex feature interactions and non-linear relationships. The findings suggest that electricity consumption in Kazakhstan may be more influenced by consistent daily industrial patterns than by residential usage that typically varies between weekdays and weekends. This research provides valuable insights for electric utility planning in Kazakhstan and demonstrates that appropriate model selection based on data characteristics is crucial for accurate load forecasting. The methodologies and findings contribute to the growing body of knowledge on AI applications in power systems management, particularly in regions with unique consumption patterns.