Secured and Smart System for Energy Management in Microgrids Using Deep Reinforcement Learning

This paper investigates the integration of Deep Reinforcement Learning (DRL) and Model Predictive Control (MPC) for optimized thermal energy management. This paper presents a novel comparative analysis of online-offline DRLMPC approaches, highlighting their respective advantages and limitations. Online DRLMPC adapts dynamically based on real-time environmental feedback, making it more effective in uncertain and fluctuating energy demand conditions. In contrast, offline DRL-MPC, trained on historical data, offers reduced computational overhead but lacks the flexibility to adjust to real-time changes. To enhance the performance of DRLMPC, the model incorporates hybrid optimization methods such as Differential Evolution (DE) and Particle Swarm Optimization (PSO), fine-tuning system parameters for improved efficiency (computational efficiency) and cost reduction. The results show that online DRLMPC outperforms offline DRL-MPC in terms of cost savings, energy efficiency, and thermal comfort, though at the cost of higher computational demands. Evaluation metrics such as cost reduction, occupant comfort levels, and computational efficiency are used to assess system performance.