Urban areas account for significant share of global electricity consumption, leading to increased energy demand, rising costs, and environmental challenges such as higher carbon emissions. Predicting electricity consumption accurately is crucial for optimizing energy usage, reducing waste, and supporting sustainable urban planning. Deep learning-based approaches, particularly convolutional neural networks (CNNs) combined with long short-term memory (LSTM) models, provide powerful solutions by leveraging large datasets to detect complex, non-linear consumption patterns. These models outperform traditional statistical methods by effectively capturing temporal dependencies, dynamic fluctuations, and spatial correlations in electricity demand. By integrating real-time data from IoT-enabled sensors and smart meters, CNN-LSTM models can enhance prediction accuracy, facilitate demand-response strategies, and improve grid stability. Additionally, incorporating transmission and distribution factors allows for a more holistic approach to energy forecasting, reducing inefficiencies and optimizing load balancing. This review provides a comprehensive overview of recent advancements in deep learning-based energy prediction, key methodologies, and their advantages over conventional techniques. However, challenges such as model generalization, computational complexity, and real-time implementation must be addressed. Future research should focus on improving model interpretability, scalability, and robustness to achieve efficient, sustainable, and cost-effective urban energy management while supporting the transition to smart grids and renewable energy integration.