Scholarlink

<p>In the context of actively addressing climate change globally and striving to achieve the “dual carbon” goal, optimizing energy management in the construction industry, especially in the high energy consuming hotel industry, is crucial. Traditional energy management strategies based on rule-based or simple model predictions are difficult to effectively address highly dynamic and uncertain challenges such as fluctuations in hotel occupancy rates, personalized comfort needs, and time of use electricity prices. Reinforcement Learning (RL), as an artificial intelligence technology that can learn long-term optimal strategies through interactive trial and error, provides new ideas for the above problems. However, existing RL research mostly focuses on residential or office buildings, and its direct application in hotel scenarios faces unique challenges such as heterogeneous state spaces, mixed action spaces, and difficulty in accurately quantifying guest satisfaction with reward functions. Therefore, this article proposes an innovative design of hotel energy management system based on reinforcement learning. Firstly, a refined Markov Decision Process (MDP) model was constructed to accurately characterize key operational characteristics of hotels, such as room occupancy status and equipment heterogeneity. Secondly, a hierarchical reinforcement learning algorithm suitable for mixed action spaces was designed, and a transfer learning mechanism was integrated to accelerate the deployment of strategies between different branches. The core algorithm is based on the Soft Actor Critic (SAC) framework, which processes mixed action spaces by selecting control modes through upper level meta strategies and executing fine control through lower level sub strategies. Finally, a high fidelity simulation environment was constructed based on real hotel data for verification. From the experimental results we can find that, the algorithm suggested by us can save energy well, reduce cost, and keep people comfortable. Take summer as an example which can reduce 29.0% of energy consumption and 31.0% of cost to achieve a satisfied comfort rating of 92.7%, while thumb, of the rule, rule of the model grease control, deep momentum control like DDPG and SAC are both very far from this satisfied level; from the array of robustness and generalization experiment of a-Buzz explosive enhancement by AWS machine learning SDK, we can see that such an algorithm possesses excellent adaptability and robustness performance to noisy interference, which makes it likely for it to be extended to the distinct overall condition of the guest house and diverse climate environment and also provides readily powerful ideas for the related upgrading roadmap of the hotel energy management.</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

Innovative design of hotel energy management system based on reinforcement learning optimization

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月15日

查看全文
  • Hsinya Hung,
  • Yingwei Wu

摘要

In the context of actively addressing climate change globally and striving to achieve the “dual carbon” goal, optimizing energy management in the construction industry, especially in the high energy consuming hotel industry, is crucial. Traditional energy management strategies based on rule-based or simple model predictions are difficult to effectively address highly dynamic and uncertain challenges such as fluctuations in hotel occupancy rates, personalized comfort needs, and time of use electricity prices. Reinforcement Learning (RL), as an artificial intelligence technology that can learn long-term optimal strategies through interactive trial and error, provides new ideas for the above problems. However, existing RL research mostly focuses on residential or office buildings, and its direct application in hotel scenarios faces unique challenges such as heterogeneous state spaces, mixed action spaces, and difficulty in accurately quantifying guest satisfaction with reward functions. Therefore, this article proposes an innovative design of hotel energy management system based on reinforcement learning. Firstly, a refined Markov Decision Process (MDP) model was constructed to accurately characterize key operational characteristics of hotels, such as room occupancy status and equipment heterogeneity. Secondly, a hierarchical reinforcement learning algorithm suitable for mixed action spaces was designed, and a transfer learning mechanism was integrated to accelerate the deployment of strategies between different branches. The core algorithm is based on the Soft Actor Critic (SAC) framework, which processes mixed action spaces by selecting control modes through upper level meta strategies and executing fine control through lower level sub strategies. Finally, a high fidelity simulation environment was constructed based on real hotel data for verification. From the experimental results we can find that, the algorithm suggested by us can save energy well, reduce cost, and keep people comfortable. Take summer as an example which can reduce 29.0% of energy consumption and 31.0% of cost to achieve a satisfied comfort rating of 92.7%, while thumb, of the rule, rule of the model grease control, deep momentum control like DDPG and SAC are both very far from this satisfied level; from the array of robustness and generalization experiment of a-Buzz explosive enhancement by AWS machine learning SDK, we can see that such an algorithm possesses excellent adaptability and robustness performance to noisy interference, which makes it likely for it to be extended to the distinct overall condition of the guest house and diverse climate environment and also provides readily powerful ideas for the related upgrading roadmap of the hotel energy management.