Abstract
The detection of refrigerant leak is critical for the effective operation of heat pump systems and vehicle maintenance. Using feature selection methods, machine learning methods, and system sensors schemes, this article optimizes the performance of refrigerant leak detection for vehicle heat pump systems that operate in cooling, heating, and series dehumidifying modes. The extremely randomized trees (EXT) model was chosen as the best model for refrigerant leak detection in heat pump systems out of 25 machine learning models, with the highest f1 score of 95.73%. The advanced feature recursive elimination strategy developed in this study is effective at increasing model accuracy, with an improvement of 1% in model diagnostic accuracy while reducing the number of features for modeling. By comparing different sensors schemes for vehicle heat pump system, this article analyzes the importance of different sensors for refrigerant leak detection. The accuracy of refrigerant leak detection in vehicles reaches 95.69% based on the optimal sensors scheme. The results demonstrate that the methods proposed in this paper not only reduce the cost of sensors used in the heat pump system, but also improve the heat pump system's performance for refrigerant leak diagnosis across three operation modes. In addition, the approach taken in this paper is broadly applicable to diagnosis problems in other fields to facilitate the application of diagnosis in the real world.
NSTL
Elsevier