Authors:
S. Revathi, S. Silvia Priscila, B. M. Praveen

Addresses:
Institute of Computer Science and Information Science, Srinivas University, Dakshina Kannada, Karnataka, India. Department of Computer Science, Faculty of Science and Humanities, SRM Institute of Science and Technology, Kattankulathur, Tamil Nadu, India. Department of Computer Science, Bharath Institute of Higher Education and Research, Chennai, Tamil Nadu, India. Institute of Engineering and Technology, Srinivas University, Dakshina Kannada, Karnataka, India.

Abstract:

Energy optimisation in smart buildings is a challenging problem that not only improves energy efficiency but also ensures occupant comfort under dynamic environmental and occupancy conditions. Conventional building management systems (BMS) and model-based predictive control (MPC) techniques are not generalised in real time due to model uncertainty, non-stationary energy consumption patterns, and the high dimensionality of sensor data in Internet of Things (IoT)-based environments. This study proposes a real-time adaptive energy management framework integrated with a Graph-Driven Long Short-Term Memory (GDLSTM) network and a Reinforcement Learning (RL)-based Multi-Objective Optimisation (MOO) mechanism to alleviate the above limitations. The GDLSTM is used to predict short-term trajectories of energy demand and comfort, accounting for spatial dependencies between building zones, and the RL agent uses Soft Actor-Critic (SAC) to learn optimal control policies that balance energy, comfort, cost, and emissions. Using a surrogate-assisted NSGA-II, the multi-objective optimiser dynamically optimises control policies that trade off conflicting objectives. Experimental evaluations on an IoT simulated Smart Building testbed with real-world datasets (ASHRAE and BEMS-Open) show that significant quantitative improvements are achieved: 18% decrease in total energy consumption, 15% cost reduction, 28% less comfort violations, and 20% emission decrease compared to baseline controllers, i.e., PID, MPC, and DDPG-based RL. 

Keywords: Smart Buildings; Energy Management; Adaptive Control; Predictive Control; Real-Time Control; Sustainable Buildings; Occupant Comfort; Energy Efficiency; Emission Reduction.

Received on: 02/12/2024, Revised on: 05/02/2025, Accepted on: 19/04/2025, Published on: 16/12/2025

DOI: 10.69888/FTSESS.2025.000554

FMDB Transactions on Sustainable Environmental Sciences, 2025 Vol. 2 No. 4, Pages: 183-199

• Views : 14
• Downloads : 6

Download PDF