Scholarlink

<p>In the current global energy depletion, building energy consumption continues to rise. Traditional energy consumption prediction methods have insufficient multi-objective coordination ability. Therefore, a building energy consumption prediction model for refined management of energy consumption in large public buildings is proposed, which integrates Prophet model and Categorical Boosting (CatBoost) model based on Stacking algorithm. The research utilizes the Prophet model to decompose trends, seasonal and holiday components, and combines it with the Mann-Kendall algorithm to accurately detect mutation points. The target encoding formula of CatBoost model is utilized to process category features, and the asynchronous successive halving algorithm is introduced to accelerate the training process. The minimum Root Mean Square Error (RMSE) reached 13.1kWh, and the maximum inference latency increased to 64.3ms. In the practical application, the shortest attenuation rate was 2.1%, the maximum RMSE increase reached 47.5%, and the difference between the peak and the measured peak was within 1%. The results indicate that the model has excellent computational efficiency and prediction accuracy, and can provide high-precision and low latency prediction tools for refined energy consumption management in typical high energy consumption scenarios such as office 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

Intelligent system for fine building energy consumption management through optimized and integrated machine learning models

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月19日

查看全文
  • Lujian Zhang,
  • Xiaoge Gao

摘要

In the current global energy depletion, building energy consumption continues to rise. Traditional energy consumption prediction methods have insufficient multi-objective coordination ability. Therefore, a building energy consumption prediction model for refined management of energy consumption in large public buildings is proposed, which integrates Prophet model and Categorical Boosting (CatBoost) model based on Stacking algorithm. The research utilizes the Prophet model to decompose trends, seasonal and holiday components, and combines it with the Mann-Kendall algorithm to accurately detect mutation points. The target encoding formula of CatBoost model is utilized to process category features, and the asynchronous successive halving algorithm is introduced to accelerate the training process. The minimum Root Mean Square Error (RMSE) reached 13.1kWh, and the maximum inference latency increased to 64.3ms. In the practical application, the shortest attenuation rate was 2.1%, the maximum RMSE increase reached 47.5%, and the difference between the peak and the measured peak was within 1%. The results indicate that the model has excellent computational efficiency and prediction accuracy, and can provide high-precision and low latency prediction tools for refined energy consumption management in typical high energy consumption scenarios such as office buildings.