Holistic optimization of air conditioning system for balanced thermal comfort and energy efficiency in smart buildings

This study proposes a novel control strategy that integrates an improved sparrow search algorithm (ISSA)-optimized long short-term memory (LSTM) network and deep Q-network (DQN) reinforcement learning, referred to as ISSA-LSTM, to optimize split air conditioning systems in rural residential buildings in Southern Shaanxi, China. Combining simulations and real-world implementations across three households, the research addresses the instability of power systems caused by cyclical air conditioning usage and irregular occupant behavior. The strategy balances energy efficiency and thermal comfort by dynamically adjusting temperature setpoints. In the fixed setpoint mode, indoor temperatures increased by 1.1–1.7 °C in summer (compared to a baseline of 26 °C) and 0.2–1.5 °C in winter (compared to 21 °C), with predicted mean vote (PMV) values maintained within the comfort range (0–0.5 in summer, −0.1 to 0.7 in winter, where PMV = 0 indicates optimal neutrality) and energy savings of 31.22% (summer) and 18.26% (winter) compared to conventional proportional integral derivative (PID) control. Under variable setpoint mode using DQN, temperatures decreased by 1.6–2.5 °C in winter and 0.5 °C in summer, yielding energy savings of 6.86% and 34.4%, respectively, while sustaining thermal comfort. The ISSA-LSTM model, surpassing traditional methods, boasts over 71.5% prediction accuracy and enables adaptive control in dynamic scenarios. This work highlights the effectiveness of hybrid AI-driven strategies in reducing grid stress and advancing sustainable building practices.