Water-chilling unit fault diagnosis method of integrating SVM mechanism

Abstract

The invention discloses a water-chilling unit fault diagnosis method of integrating an SVM mechanism. According to the method, unit no-fault and fault running history data are obtained through sensorsin the on-site water-chilling unit, and are used for no-fault and various-fault DW-SVDD model and full-fault SVM model training; a trained no-fault DW-SVDD model is applied for online fault detection, the on-site water-chilling unit is normal if a condition is met; otherwise, DW-SVDD models are used for on-line diagnosis, a type of fault is diagnosed if the actually measured data only meet a condition of one fault model therein; a full-fault SVM model is used for re-diagnosis for determining a fault type thereof if the actually measured data meet conditions of more than two fault models therein, that is, diagnosis results are confused; and a new type of fault is diagnosed if the actually measured data does not meet each fault model. The method effectively overcomes main limitation existing in current fault diagnosis stages, and has higher diagnosis correctness and accuracy.

Description

technical field [0001] The invention belongs to the technical field of fault diagnosis of water chillers in an air-conditioning system, and in particular relates to a fault diagnosis method for chillers integrating a support vector machine (SVM) mechanism. Background technique [0002] The chiller is the main energy-consuming equipment in the air conditioning system. Therefore, through the fault diagnosis method, the performance degradation during the operation of the chiller can be found in time, and the operation and maintenance can be carried out in time to ensure that the chiller can operate normally at a higher energy efficiency level. It not only reduces operating energy consumption, but also saves operating and maintenance costs. [0003] At present, the mainstream methods of chiller fault detection and diagnosis are methods based on historical data, including artificial neural network, multiple linear regression, fuzzy logic and support vector data description, etc. ...

AI Technical Summary

Problems solved by technology

[0004] 1) Methods such as artificial neural network, multiple linear regression, and fuzzy logic have a large demand for samples and low diagnostic accuracy;

[0005] 2) When the method of support vector data description is used for fault detection, the false alarm rate is high, the diagnostic results are easily confused, and the misdiagnosis rate is high.

[0006] Due to the difficulty of obtaining various fault data in the computer room, traditional one-class classification methods (such as SVDD) require less training data samples, but the accuracy of fault detection and diagnosis is low, mainly because the data density distribution is not considered. Therefore, on the basis of considering the density distribution, this research group proposed a classification method based on density-weighted support vector data description (DW-SVDD), which reduces the false alarm rate in the detection stage, but the diagnosis results are confused , high rate of misdiagnosis

Images