AI-driven energy optimization enhancing efficiency in urban environments with hybrid machine learning models

Accurate forecasting of electricity consumption is essential for sustainable urban planning, particularly in fast-growing cities like Tehran. Conventional models often fail to adequately capture the intricate relationships between environmental factors and energy demand. To overcome these limitations, this study applies advanced AI techniques such as Neural Networks, Random Forest Regression, and Gradient Boosting, using a comprehensive dataset (2000–2022) that integrates meteorological, environmental, and fuel consumption variables to enhance predictive performance. Random Forest Regression achieved the highest accuracy, with an R 2 0.9835 and MSE of 0.0165, explaining 98.35 % of the variation in electricity consumption. Feature engineering substantially improved model accuracy, highlighting temperature variables (T2M, T2M_MAX, T2M_MIN) and fuel consumption as the most influential predictors. Correlation analysis revealed strong associations between environmental factors and electricity demand. Using Sequential Least Squares Programming (SLSQP) optimization, the study determined conditions that reduced electricity consumption to 1.09 million kWh. These findings highlight the value of AI models in enhancing forecasting accuracy and supporting efficient energy planning. Ensemble learning and optimization methods strengthen sustainable energy management. However, reliance on historical data and neglect of socio-economic factors may constrain the models’ adaptability and predictive power. Moreover, the complexity of AI models presents interpretability challenges, requiring additional efforts to align outputs with policy-making needs. Leveraging AI and data-driven methods, this study offers actionable insights for policymakers to optimize energy use and curb emissions in urban settings like Tehran. Future research should incorporate socio-economic variables and hybrid models to enhance predictive reliability and practical relevance. • Random Forest Regr achieved R 2 of 0.9835 and MSE of 0.0165, accurately capturing electricity demand patterns in Tehran. • Add variables such as temperature, wind energy potential, and fuel consumption significantly enhanced model accuracy, particularly for capturing both standard and peak demand periods. • SLSQP optimization cut electricity use to 1.09 million kWh boosting energy efficiency. • RF, GBM, and DT models outperformed traditional regression and standalone deep learning models, which struggled with overfitting.