Intelligent HVAC Control in Residential Buildings: A Systematic Review of Advanced Techniques and AI Applications

This systematic review on intelligent HVAC systems for residential buildings focuses on advanced control techniques and AI applications. Model Predictive Control (MPC) is the most common method (≈40% of studies), achieving 15–20% energy savings and 10–30% peak demand reduction. Deep reinforcement learning (DRL) offers a model-free alter-native, reducing energy costs by 15% and comfort violations by up to 98%. Neural networks (LSTM, CNN-BiLSTM, attention mechanisms) significantly improve load pre-diction and thermal comfort modelling, with fusion models boosting accuracy by 66–85%. Comprehensive AI-based systems deliver 22–44% energy savings and 22–86% comfort improvements. Performance varies by climate, building type, and baseline; field trials show lower but more reliable savings than simulations. Hybrid MPC–ML approaches are emerging as best practice. Barriers include model complexity, computational demands, limited training data, and integration with legacy systems. Occupancy-aware strategies save 19–45% energy, while intelligent thermal storage can raise solar fractions from 11% to 61%. Overall, intelligent HVAC control is technically feasible and economically advantageous, but success depends on accurate modelling, tailored control strategies, and robust han-dling of occupancy uncertainty.