Scholarlink

<p>Home energy management system (HEMS), combines with renewable energy and energy storage systems (ESS), has emerged as a promising technique to reduce the operation and carbon emission costs. However, the uncertainties of both the generation of renewable energy and the energy consumption in user side, make it challenging to design an efficient demand response strategy in HEMS. In this paper, we introduce a HEMS that unifies rooftop photovoltaic (PV), ESS, electric vehicles, and household appliances to enhance the performance of demand response strategy. The ESS is connected with the PV, and they participate in the system operation as a whole. We propose a dual-agent deep reinforcement learning (DADRL) approach, combining deep deterministic policy gradient (DDPG) and deep Q-network (DQN) methodologies to refine DR mechanisms. The DDPG algorithm optimizes ESS operation in light of PV forecasting errors, while the DQN algorithm schedules household appliances to maximize user satisfaction while minimize energy cost. Simulation results indicate that the DADRL approach mitigates PV forecasting errors substantially. It reduces the electricity costs, and increases occupant satisfaction of users. The efficacy of integrating DR strategies with renewable energy is demonstrated to manage the uncertainties in both the energy supply and consumption sides.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Demand response for home energy management systems: a novel dual-agent DRL approach

  • 期刊论文

  • 发表日期：

    2025年7月8日

查看全文
  • Weiming Zheng,
  • Runyi Pi,
  • Xiaoqing Zhong,
  • Chao Yang

摘要

Home energy management system (HEMS), combines with renewable energy and energy storage systems (ESS), has emerged as a promising technique to reduce the operation and carbon emission costs. However, the uncertainties of both the generation of renewable energy and the energy consumption in user side, make it challenging to design an efficient demand response strategy in HEMS. In this paper, we introduce a HEMS that unifies rooftop photovoltaic (PV), ESS, electric vehicles, and household appliances to enhance the performance of demand response strategy. The ESS is connected with the PV, and they participate in the system operation as a whole. We propose a dual-agent deep reinforcement learning (DADRL) approach, combining deep deterministic policy gradient (DDPG) and deep Q-network (DQN) methodologies to refine DR mechanisms. The DDPG algorithm optimizes ESS operation in light of PV forecasting errors, while the DQN algorithm schedules household appliances to maximize user satisfaction while minimize energy cost. Simulation results indicate that the DADRL approach mitigates PV forecasting errors substantially. It reduces the electricity costs, and increases occupant satisfaction of users. The efficacy of integrating DR strategies with renewable energy is demonstrated to manage the uncertainties in both the energy supply and consumption sides.