Research on Typical Fault Diagnosis of Chillers Based on Machine Learning

Chiller fault diagnosis technology can effectively predict failure risks, prevent catastrophic failures, and improve system reliability, thereby playing a crucial role in ensuring the safe and efficient operation of chillers. A comparative analysis of diagnostic accuracy between the Random Forest (RF) algorithm and Decision Tree (DT) algorithm shows that the RF algorithm achieves a diagnostic accuracy rate 2.74% higher than that of the DT algorithm. Using a confusion matrix, the diagnostic principles of three typical faults with similar characteristics are discussed, and a stepwise diagnostic model based on the Random Forest algorithm is proposed. By optimizing feature selection, tuning model hyperparameters, and expanding the data sample size, the diagnostic accuracy of the proposed model is further improved, reaching 97%.