Publication

Abstract : In today's energy-conscious world, efficient energy management has become a critical challenge. Buildings, being one of the major energy consumers' usages of light source, humidity control irrespective to surrounding environment conditions and presence of human, requires innovative solutions to optimize energy consumption. The proposed study presents a EMDRL (Energy Management Deep Reinforcement Learning), an energy management system for room optimization that makes use of Deep Reinforcement Learning (DRL) approaches notably Deep Q-Learning, dynamic decision making of a DRL agent with environment interaction. Deep Reinforcement Learning extends the principles of RL by incorporating Deep Neural Networks to learn optimal behaviors. This includes variations of DQN algorithms, such as Deep Q-Network (DQN), Double Deep Q-Network (DDQN), Dueling Q-Network (Dueling DQN) and its hyperparameter tuning and comparative analysis of the three algorithms and an evaluation of their performance. The comparative analysis has contributed valuable insights, providing a viable resolution to the problems associated with modern building energy management and cost efficiency in which Dueling DQN our novel model's agent that has contributed a significant performance of 136%, 144% more reward gaining and less energy-efficient by saving cost compared to Double DQN and traditional DQN respectively. © 2024 IEEE.