Analysis of a city's sustainable energy potential using artificial intelligence: The case of Herat, Afghanistan

This doctoral thesis investigates the electricity consumption and sustainable energy potential of Herat City using artificial intelligence applications. To achieve this, energy consumption predicting models were developed, and analyses supporting sustainable energy integration and efficiency were conducted. Advanced machine learning methodologies were applied using three years of hourly electricity consumption and meteorological data from Herat City. Data preprocessing and correlation analysis revealed strong correlations between dew point and energy consumption, and inverse correlations with solar radiation. A comparative analysis was performed on Support Vector Regression (SVR), Random Forest (RF), Neural Networks (NN), and Long Short-Term Memory (LSTM) networks. Among these models, LSTM (Model 12) demonstrated superior predicting performance, effectively capturing complex temporal dependencies in energy consumption. This model achieved a test R2 score of 0.891 and a Mean Squared Error (MSE) of 18.16. Furthermore, this thesis analyzes Herat City's potential for sustainable energy transition and concludes that the increasing demand can be met through solar and wind energy integration. The findings of this thesis also highlight the importance of strategic planning and investments for renewable energy infrastructure. Thus, this study contributes to the field of sustainable energy research by demonstrating the effectiveness of LSTM networks in predicting energy consumption in data-constrained environments. The results of this thesis provide data-driven sustainable energy solution insights for Herat City.