Refrigerant leakage fault detection and diagnosis based on extreme gradient boosting and random forest algorithm

Wu Kongrui, Han Hua, Ren Zhengxiong, Gao Yu, Jiang Songxuan, Yang Yuting


Aiming at the situation where the normal data in the chiller unit exceeds the fault data and the most common refrigerant leakage fault in the refrigeration system, this paper uses the extreme gradient boosting (XGBoost) algorithm to establish the fault detection model, and the random forest (RF) algorithm to establish the fault diagnosis model, and studies the influence of the detection threshold change on the detection model and the comparison of the diagnostic model with and without normal sample training. The results show that when the detection threshold is set to 0.99, most of the fault samples can be detected, the false alarm rate is low, and the diagnostic model trained only by the fault data makes the best overall performance, which can maximize the advantages of the detection model and diagnostic model.