محمدحسين كرباسي

With the increasing growth of energy deman‎d in the building sector an‎d the expansion of renewable energy resources, home energy management systems play a vital role in optimizing energy consumption an‎d reducing costs. However, challenges such as uncertainty in solar power generation, dynamic fluctuations in electricity prices, an‎d the complexity of simultaneously controlling multiple devices have limited the effectiveness of traditional model-based control methods. To address these challenges, this study proposes a comprehensive framework based on deep reinforcement learning using the Deep Deterministic Policy Gradient (DDPG) algorithm. The proposed approach is capable of managing integrated energy resources in a continuous action space without requiring an accurate physical model of the building. In this model, three key components—namely the heating, ventilation, an‎d air conditioning (HVAC) system, the energy storage system (ESS), an‎d a smart water pump—are controlled simultaneously. The main innovation of this research lies in incorporating the water pump as a flexible load an‎d designing a multi-objective reward function that dynamically balances conflicting goals, including minimizing energy costs, maintaining thermal comfort, an‎d extending equipment lifespan. The agent training process is conducted in a simulated environment using real-world data an‎d in the presence of environmental noise. Simulation results demonstrate that the proposed model exhibits stable an‎d convergent performance, achieving up to a 25.4% reduction in total energy cost. At the same time, the system successfully maintains indoor temperature within the desired comfort range for 99.2% of the time. Furthermore, the learned policies lead to conservative battery management behavior—reducing degradation—an‎d enable approximately 27% of the required energy to be supplied through selling surplus solar power to the grid. Overall, the findings highlight the superiority an‎d robustness of the DDPG-based approach in han‎dling uncertainties an‎d environmental noise compared to previous methods.

مديريت انرژي خانگي , يادگيري تقويتي عميق , الگوريتم DDPG , خانه هوشمند , بهينه‌سازي چندهدفه , عدم قطعيت

Home Energy Management , Deep Reinforcement Learning , DDPG Algorithm , Smart Home , Multi-objective Optimization , Uncertainty

Mohammadhosein Karbasi

DR.Mohammadreza Rasouli