Saudi Cultural Missions Theses & Dissertations
Permanent URI for this communityhttps://drepo.sdl.edu.sa/handle/20.500.14154/10
Browse
2 results
Search Results
Item Restricted Evaluating the Effectiveness of Existing AI Models in Energy Management for Smart Facilities and Buildings(Saudi Digital Library, 2025) Aldawsari, Abdulrahman; Morgan, PeterThis project evaluates the practical effectiveness of existing artificial intelligence (AI) models used in energy management systems for smart buildings and microgrids. While the academic literature is rich in high-performing algorithms, little is known about how these models function under real-world constraints such as data availability, system integration, and operator interpretability. The research focuses on four main AI model types: deep learning models such as Long Short-Term Memory (LSTM) and Gated Recurrent Units (GRU); tree-based models including Random Forest (RF) and Gradient Boosted Trees (GBT); hybrid models combining convolutional neural networks (CNN) and support vector regression (SVR); and reinforcement learning approaches, particularly Proximal Policy Optimisation (PPO). A structured evaluation framework was developed using three pillars: technical performance, operational feasibility, and deployment readiness. Each model was assessed using peer-reviewed results and case studies, with comparative analysis across forecasting accuracy, training demands, interpretability, and integration ease. The findings revealed that deep learning models, particularly LSTM and GRU, excelled in forecasting accuracy but were resource-intensive and opaque to non-specialist users. Tree-based models such as RF offered greater transparency and were easier to deploy but had lower accuracy in complex, time-dependent scenarios. Hybrid models demonstrated the highest accuracy but required significant tuning and maintenance. PPO-based models were effective in dynamic systems like microgrids but presented challenges with explainability and reward design. Federated learning approaches showed promise in decentralised or privacy-sensitive environments, although the results were mixed and highly context-dependent. Key deployment barriers include data quality gaps, limited technical expertise, and poor interoperability with legacy building management systems. Case studies reinforce the view that no model is universally optimal; effectiveness depends on how well a model aligns with the operational environment. For example, interpretable models may be more suitable in public-sector buildings, while advanced reinforcement learning may be better suited to complex, high-investment infrastructure. The study concludes that successful adoption of AI in energy management requires more than technical optimisation. It demands models that are accurate, explainable, and compatible with the real conditions of the buildings they serve. Recommendations include selecting models based on a balance of accuracy and interpretability, planning for model retraining, addressing integration barriers early, and investing in region-specific validation to ensure broader applicability.7 0Item Restricted Enhancing Stock Price Prediction Using Machine Learning Models: A Comparative Study of SVM, LSTM, and GRU(University College London, 2024-08) AlMohamdy, Razan; Andrea, DucciThis study evaluates the effectiveness of three machine learning models—Support Vector Machine (SVM), Long Short-Term Memory (LSTM) networks, and Gated Recurrent Units (GRU)—in predicting the stock prices of Saudi Aramco. Using historical stock price data and technical indicators, the models were assessed based on their accuracy in both long-term and short-term predictions. The findings reveal that LSTM and GRU significantly outperform SVM, with LSTM showing superior performance in capturing long-term dependencies and GRU offering a balance between accuracy and computational efficiency. Specifically, LSTM achieved a Root Mean Squared Error (RMSE) of 0.0516 and a Mean Absolute Error (MAE) of 0.0323, while GRU recorded an RMSE of 0.0539 and an MAE of 0.0234. In contrast, SVM exhibited a much higher RMSE of 0.1712 and an MAE of 0.1079, indicating its struggles with market volatility. The 30-day prediction analysis further highlighted the strengths of LSTM and GRU in short-term forecasting, with both models maintaining an R² value above 0.993, while SVM lagged behind at 0.9332. Despite their advantages, the study identified limitations such as the exclusion of external economic factors and the models' varying effectiveness across different time horizons. These findings contribute to the growing field of financial forecasting, offering practical insights for investors and analysts on model selection. Future research is recommended to incorporate broader economic indicators, explore cross-market validation, and enhance the models' responsiveness to short-term market fluctuations.31 0