Multi-scale energy forecasting in smart households: Leveraging AI and remotely sensed climate data for sustainable energy management

• Developed a multi-scale AI framework using global climate and solar datasets. • Proposed a hybrid SVR–MLP model that outperforms ML/DL benchmarks. • Demonstrated adaptability of forecasting across multiple temporal scales. • Delivered a generalizable approach beyond localized datasets. • Emphasized the role of occupant-related data for human-centric forecasting. Accurate energy consumption forecasting is essential for optimizing energy use and promoting sustainability in both urban environments and smart households. This study presents an AI-driven multi-scale forecasting framework that integrates global climate and solar energy datasets from NASA and ESA with smart household consumption data. Four models, Support Vector Regression, Extreme Gradient Boosting, Long Short-Term Memory, and a hybrid Support Vector Regression - Multi-Layer Perceptron (SVR-MLP) were evaluated across multiple temporal resolutions (hourly, 3-hourly, 6-hourly, 12-hourly, and daily). The hybrid SVR–MLP consistently outperformed the other approaches, while variations across households revealed distinct consumption behaviors, underscoring the influence of occupant-specific factors on forecasting accuracy. By integrating remote sensing data with AI-based models, the proposed framework offers a flexible and scalable solution for energy monitoring and optimization, supporting eco-efficiency and data-driven sustainable energy management.