Scholarlink

Fault detection and diagnosis in commercial buildings have the potential to yield substantial energy savings. The term “light commercial building” typically refers to commercial buildings that have a surface area of less than 2350 square meters and no more than six stories. These buildings typically rely on relatively simpler HVAC systems that have similar configurations, which facilitates the development of scalable and replicable fault detection approaches. This paper introduces three inverse modeling approaches to establish a baseline for detecting anomalies in Air Handling Unit (AHU) systems within these light commercial buildings. The study aims to characterize AHU thermal loads using optimization techniques: artificial neural networks (ANNs), genetic algorithms (GA), and Physics-based neural networks (PBNN). Anomalies in energy consumption are then identified as deviations from this baseline. AHU data from a light commercial building in Montreal, Canada, is employed for validation. Results indicate superior performance of PBNN in baseline establishment compared to optimization methods. Detected anomalies using PBNN can aid in investigating programming logic faults following ASHRAE Guideline 36. The study underscores the importance of robust anomaly detection methods in enhancing the operational efficiency of AHU systems, particularly in light commercial buildings.

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

Physics-Based Inverse Model Anomaly Detection in Light Commercial Buildings’ AHU Systems

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Milad Babadi Soultanzadeh,
  • Mazdak Nik-Bakht,
  • Mohamed M. Ouf,
  • Pierre Paquette,
  • Steve Lupien

摘要

Fault detection and diagnosis in commercial buildings have the potential to yield substantial energy savings. The term “light commercial building” typically refers to commercial buildings that have a surface area of less than 2350 square meters and no more than six stories. These buildings typically rely on relatively simpler HVAC systems that have similar configurations, which facilitates the development of scalable and replicable fault detection approaches. This paper introduces three inverse modeling approaches to establish a baseline for detecting anomalies in Air Handling Unit (AHU) systems within these light commercial buildings. The study aims to characterize AHU thermal loads using optimization techniques: artificial neural networks (ANNs), genetic algorithms (GA), and Physics-based neural networks (PBNN). Anomalies in energy consumption are then identified as deviations from this baseline. AHU data from a light commercial building in Montreal, Canada, is employed for validation. Results indicate superior performance of PBNN in baseline establishment compared to optimization methods. Detected anomalies using PBNN can aid in investigating programming logic faults following ASHRAE Guideline 36. The study underscores the importance of robust anomaly detection methods in enhancing the operational efficiency of AHU systems, particularly in light commercial buildings.