Research on label propagation based on clustering and semi-supervised learning under limited fault data of air conditioning system

Abstract

Heat pump systems may experience various faults under complex operating conditions, making fault diagnosis crucial for ensuring reliable operation and effective equipment management. At present, related research in this area is highly dependent on labeled data, while few studies focus on using unsupervised learning to process unlabeled data. However, it is difficult to obtain real and labeled fault data, which is an important factor limiting the development of fault detection and diagnosis technology. Therefore, two label propagation strategies are proposed in this study, including the clustering-based model and the model combined with semi-supervised learning. Through the proposed method, limited labeled data can be utilized to expand the dataset. In addition, the effects of different parameters on the accuracy performance of the models are compared. Finally, the two models are optimized by using ensemble clustering, which significantly improves the accuracy of label propagation. Specifically, after optimization with ensemble clustering, the overall accuracy of clustering-based and semi-supervised label propagation models can reach up to 98.0 % and 94.5 %, respectively. In terms of label propagation accuracy for each fault type, both models avoid serious misjudgment of minority class data after optimization, and the label propagation accuracy of other types has also been effectively improved. The research results show that the method proposed in this study can effectively address the problem of low utilization of unlabeled data, thereby improving the fault diagnosis performance of the machine learning model of air conditioning system.