Abstract:Faults in HVAC systems degrade thermal comfort and energy efficiency in buildings and have received significant attention from the research community, with data driven methods gaining in popularity. Yet the lack of labeled data, such as normal versus faulty operational status, has slowed the application of machine learning to HVAC systems. In addition, for any particular building, there may be an insufficient number of observed faults over a reasonable amount of time for training. To overcome these challenges, we present a transfer methodology for a novel Bayesian classifier designed to distinguish between normal operations and faulty operations. The key is to train this classifier on a building with a large amount of sensor and fault data (for example, via simulation or standard test data) then transfer the classifier to a new building using a small amount of normal operations data from the new building. We demonstrate a proof-of-concept for transferring a classifier between architecturally similar buildings in different climates and show few samples are required to maintain classification precision and recall.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: In heating, ventilation, and air - conditioning (HVAC) systems, the application of machine - learning methods is limited due to the lack of labeled data (such as data on normal and faulty operating states). Specifically, for a particular building, the number of faults observed within a reasonable time may not be sufficient for effective training. To address this challenge, the authors propose a transfer - learning method for transferring a new - type Bayesian classifier from a building with a large number of sensors and fault data to a new building, thereby achieving effective discrimination between normal and faulty operating states. ### Specific Problem Description 1. **Lack of Labeled Data**: Fault detection in HVAC systems depends on labeled data, but obtaining this data is very difficult and time - consuming. Even if a building is equipped with a large number of sensors, it is very complicated to correctly label whether each data point comes from a fault or normal operation. 2. **Insufficient Fault Samples for a Specific Building**: Even with sufficient sensor data, the number of faults that may occur in a specific building within a period of time is limited, and it is difficult to collect enough fault samples for effective training. 3. **Requirement for Transfer Learning**: To overcome the above problems, a method is needed that can transfer existing knowledge between different buildings to reduce the need for a large amount of labeled data in new buildings. ### Solution Overview The authors propose a Bayesian classifier based on transfer learning to solve the above problems through the following steps: 1. **Model Training**: Train the classifier on a building with a large number of sensors and fault data. These data can come from actual measurements or simulations. 2. **Model Transfer**: Use a small amount of normal - operation data from the new building to transfer the trained classifier to the new building. The key is to use the rich data of the source building to constrain the degrees of freedom of the target - building model, thereby improving the transfer efficiency. 3. **Verification and Application**: Verify the effectiveness of this method through experiments and show that when transferring between similar buildings under different climatic conditions, only a small number of samples are required to maintain the precision and recall rate of the classifier. ### Mathematical Expressions The mathematical formulas involved in the paper include, but are not limited to: - **Estimation of the State - Transition Matrix**: \[ \hat{A}_1=\arg\min_W \|W S^{(1)} - X^{(1)}\|_2^2 \] where \(S^{(1)}\) and \(X^{(1)}\) are the state - input and output data of the source building respectively. - **Weighted Least - Squares Method**: \[ \hat{A}_2=\arg\min_W \|W [S^{(1)}, S^{(2)}] - [X^{(1)}, X^{(2)}]\|_2^2 \] where \(S^{(2)}\) and \(X^{(2)}\) are the data of the target building, and weights are assigned to the data of the source building and the target building respectively. - **Classification Rule**: \[ 0 \gtrless \log[P(x_{t + 1}|A, s_t)]-\log[P(x_{t + 1}|\tilde{A}, s_t)] \] where \(P(x_{t + 1}|A, s_t)\) and \(P(x_{t + 1}|\tilde{A}, s_t)\) represent the probabilities of generating the observed value \(x_{t+1}\) in the normal state and the faulty state respectively. Through these methods, the authors show how to effectively transfer the fault - detection model of HVAC systems from one building to another, thereby improving the efficiency and accuracy of fault detection.

Transfer Learning for HVAC System Fault Detection

Model-based Fault Detection and Diagnosis for HVAC.

A novel image-based transfer learning framework for cross-domain HVAC fault diagnosis: From multi-source data integration to knowledge sharing strategies

Development of an FDD model for an existing building using transfer learning

Transfer learning-based strategies for fault diagnosis in building energy systems

A thermodynamic-law-integrated deep learning method for high-dimensional sensor fault detection in diverse complex HVAC systems

Advancing Fault Detection in HVAC Systems: Unifying Gramian Angular Field and 2D Deep Convolutional Neural Networks for Enhanced Performance

One for Many: Transfer Learning for Building HVAC Control

Fault detection and diagnosis in light commercial buildings' HVAC systems: A comprehensive framework, application, and performance evaluation

Comparative study on deep transfer learning strategies for cross-system and cross-operation-condition building energy systems fault diagnosis

Integrating Active Learning and Semi-Supervised Learning for Improved Data-Driven HVAC Fault Diagnosis Performance

An Experimental HVAC Faults Data Generation and Detection Using One-Dimensional Convolutional Neural Networks

Learning-based Framework for Sensor Fault-Tolerant Building HVAC Control with Model-assisted Learning

Transfer learning with deep neural networks for model predictive control of HVAC and natural ventilation in smart buildings

Transfer Learning in the Transformer Model for Thermal Comfort Prediction: A Case of Limited Data

Fault diagnosis of HVAC system with imbalanced data using multi-scale convolution composite neural network

Transfer Learning for Thermal Comfort Prediction in Multiple Cities

Experimental study on performance assessments of HVAC cross-domain fault diagnosis methods oriented to incomplete data problems

Fast Adaptation of Thermal Dynamics Model for Predictive Control of HVAC and Natural Ventilation Using Transfer Learning with Deep Neural Networks.

Leveraging graph convolutional networks for semi-supervised fault diagnosis of HVAC systems in data-scarce contexts

Integrated building fault detection and diagnosis using data modeling and Bayesian networks