Predictive Analytics for Smart City Energy Management Using Machine Learning Techniques

Abstract

Accurate short-term electricity demand forecasting is a critical requirement for data-driven planning and operational support in smart city energy systems. Urban energy consumption exhibits strong temporal dependencies, nonlinear behavior, and sensitivity to exogenous factors such as weather and human activity patterns, which limit the effectiveness of traditional linear forecasting methods. This study examines the application of machine learning techniques for short-term energy demand prediction, aiming to enhance forecast accuracy and interpretability in smart city energy management contexts. A comparative modeling framework is developed, using linear regression as a baseline and Random Forest, Gradient Boosting, and Long Short-Term Memory (LSTM) networks as advanced learners. The models are trained on historical load data enriched with temporal and meteorological features and evaluated using Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and the coefficient of determination (R²). The results demonstrate that nonlinear and sequence-based models consistently outperform linear baselines, with the LSTM achieving the lowest error and highest explanatory power by effectively capturing temporal dynamics in urban energy demand. Feature-importance analysis across models reveals that recent historical load is the dominant predictor, complemented by calendar effects and weather variables, underscoring the combined influence of behavioral regularity and environmental conditions. These findings indicate that machine learning-based forecasting models provide a more accurate and informative basis for short-term energy planning in smart cities, supporting data-driven decision-making while highlighting the need for robustness and interpretability in real-world deployments.