Scholarlink

<p>This paper presents an implementation of a Q-Deep Network (DQN) for energy management in smart homes using deep reinforcement learning (DRL) techniques, leveraging the CLEAN_House dataset. The goal is to optimize energy consumption while maintaining comfort and functionality. The study evaluates three different DQN architectures, with DQN2 demonstrating superior performance. The training process, which includes data normalization, model definition, and a detailed exploration–exploitation strategy, is systematically documented. Results indicate that the model effectively learns to minimize energy usage, as evidenced by increased total rewards and reduced losses over episodes. Despite some fluctuations in performance, the overall trend suggests significant potential for DRL applications in real-world energy management. Continuous learning and more sophisticated techniques are recommended for handling dynamic conditions. This work highlights the efficacy of DQN in achieving energy savings in smart homes, paving the way for scalable and adaptable energy management solutions. The novelty of this study lies in its implementation of a DQN for real-time energy management in smart homes, demonstrating its potential to learn and optimize energy consumption autonomously.</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

Optimizing energy management in smart homes with Q-deep networks: a deep reinforcement learning approach for sustainable consumption

  • 期刊论文

  • 发表日期：

    2025年8月6日

查看全文
  • Mohammadreza Ganjian,
  • Mohammad Tabrizian,
  • Nasser Khodabakhshi,
  • Meqdad Ansarian

摘要

This paper presents an implementation of a Q-Deep Network (DQN) for energy management in smart homes using deep reinforcement learning (DRL) techniques, leveraging the CLEAN_House dataset. The goal is to optimize energy consumption while maintaining comfort and functionality. The study evaluates three different DQN architectures, with DQN2 demonstrating superior performance. The training process, which includes data normalization, model definition, and a detailed exploration–exploitation strategy, is systematically documented. Results indicate that the model effectively learns to minimize energy usage, as evidenced by increased total rewards and reduced losses over episodes. Despite some fluctuations in performance, the overall trend suggests significant potential for DRL applications in real-world energy management. Continuous learning and more sophisticated techniques are recommended for handling dynamic conditions. This work highlights the efficacy of DQN in achieving energy savings in smart homes, paving the way for scalable and adaptable energy management solutions. The novelty of this study lies in its implementation of a DQN for real-time energy management in smart homes, demonstrating its potential to learn and optimize energy consumption autonomously.