AI-optimized energy and comfort control for carbon-neutral tropical buildings

초록

Cooling energy consumption constitutes a significant portion of total energy use in buildings located in hot and humid climates. This paper presents an AI-driven integrated strategy to optimize energy efficiency and indoor comfort while reducing carbon emissions in Kuala Lumpur, Malaysia. The key innovation differentiating this approach from existing studies lies in the integration of Multi-Input Bottleneck Architecture (MIBA)-based Long Short-Term Memory models, 5 min high-resolution real-time data processing, and Model Predictive Control systems incorporating adaptive comfort models. Unlike existing AI-based HVAC research that relies on hourly control or simple neural networks, this study utilizes environmental, control, and energy data sampled at 5 min intervals to forecast indoor conditions and optimize Air Conditioning and Mechanical Ventilation control strategies. Through Walk-Forward Validation across 21 stages, the energy consumption prediction model achieves high accuracy (R2> 0.88). The optimization algorithm reduces daily energy consumption from 10,306 kWh to 9,864 kWh and peak usage from 84 kWh to 82 kWh. Over one-month simulation, total energy savings reached 12,871 kWh, CO2 emissions decreased by 5,913 kg, and cost savings amounted to, RM 6,304. Twenty-four-hour empirical validation confirmed simulation accuracy, demonstrating consistency between theoretical modeling and actual implementation. This study provides a practical solution addressing the comfort-efficiency trade-off problem and offers innovative building management approaches for achieving sustainable energy usage and contributing to long-term environmental sustainability in hot and humid climates. © 2025 The Authors

키워드