Scholarlink

<p>There is a natural contradiction between energy efficiency and comfort in building HVAC systems. This paper proposes an Augmented Intelligence Proximal Policy Optimization (AuI-PPO) control method that combines human guidance. In system modeling, indoor and outdoor temperatures, electricity prices, and human intervention signals are introduced as states, and actions are defined as setting temperature adjustments. The reward function integrates energy consumption costs and PMV comfort. The experiment is based on the TRNSYS simulation platform and compares different building areas (60, 120, 180&#xa0;m<sup>2</sup>) with climatic conditions. The results indicate that AuI-PPO outperforms PPO, TD3, HI-DDPG, and DDPG in terms of convergence speed and stability, with a final R<sup>2</sup> of over 0.95 and achieving the lowest average PMV and energy consumption load. Sensitivity analysis shows that the setting with a learning rate of 0.001, a hidden layer of 512, and τ = 1e-4 yields the best results. Overall, AuI-PPO can effectively reduce costly exploration, strike a balance between energy efficiency and comfort, and provide a feasible solution for optimizing energy efficiency in smart buildings.</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

Energy saving optimization method for building HVAC based on PPO and human guidance

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月25日

查看全文
  • Xuan Yu,
  • Wenqing Cai

摘要

There is a natural contradiction between energy efficiency and comfort in building HVAC systems. This paper proposes an Augmented Intelligence Proximal Policy Optimization (AuI-PPO) control method that combines human guidance. In system modeling, indoor and outdoor temperatures, electricity prices, and human intervention signals are introduced as states, and actions are defined as setting temperature adjustments. The reward function integrates energy consumption costs and PMV comfort. The experiment is based on the TRNSYS simulation platform and compares different building areas (60, 120, 180 m2) with climatic conditions. The results indicate that AuI-PPO outperforms PPO, TD3, HI-DDPG, and DDPG in terms of convergence speed and stability, with a final R2 of over 0.95 and achieving the lowest average PMV and energy consumption load. Sensitivity analysis shows that the setting with a learning rate of 0.001, a hidden layer of 512, and τ = 1e-4 yields the best results. Overall, AuI-PPO can effectively reduce costly exploration, strike a balance between energy efficiency and comfort, and provide a feasible solution for optimizing energy efficiency in smart buildings.