RESEARCH ON EVALUATION OF MODELS FOR CHILLER FAULT DETECTION AND DIAGNOSIS STRATEGY

Abstract

An accurate parameter model plays a key role in improving the accuracy rate of fault detection and diagnosis in chiller systems. Therefore, in this study, 3 methods of MLR, GRNN and RBFNN have been compared and evaluated as a parametric model to model the operating characteristics of the chiller. Two statistical indexes, R² and RMSE, are used as model evaluation criteria at the model training stage. Then, combined with the t-test method and the diagnostic rule to study, survey and evaluate the diagnostic detection ability of the 3 models. The commonly experimental data set for the research direction of detection and diagnosis of faults in the chiller system of ASHRAE RP-1043 was used in this study. The study conducted to evaluate 3 models with 3 typical cases: "Chiller works normally" and 2 typical faults in chiller system "Refrigerant leakage", "Condenser fouling". The results of the study show that, RBFNN and GRNN are very practical and highly accurate strategy.