Welcome to the Feature Selection for Chiller FDD project page!


Motivation of the current work:

Chiller fault detection and diagnosis is a well studied work in the literature. Various solutions were proposed. Among them, different feature subsets were used in different works, cause mass confusion. In this study, we tend to adopt the cost-sensitive learning method and select the optimal feature subsets using back-traced sequential forward feature selection (BT-SFS) algorithm.


Preprocessed chiller dataset (From ASHRAE project 1043-RP)

The data was inherited from ASHRAE project coded 1043-RP, which including 1 normal set of data, 7 faults and 4-5 sever levels for each fault. Therefore, there are in total 30 labels with 65 features:

TWE_set TEI TWEI TEO TWEO TCI TWCI TCO TWCO TSI TSO TBI TBO CondTons CoolingTons SharedCondTon
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
CondEnergyBalance EvapTons SharedEvapTons BuildingTons EvapEnergyBalance kW COP kW/Ton FWC FWE TEA TCA TRE PRE TRC PRC
17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32
TRC_sub T_suc Tsh_suc TR_dis Tsh_dis P_lift Amps RLA% HeatBalance(kW) HeatBalance% Tolerance% UnitStatus ActiveFault TO_sump TO_feed PO_feed
33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48
PO_net TWCD TWED VSS VSL VH VM VC VE VW TWI TWO THI THO FWW FWH FWB
49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

After possible preprocessing (including normalization, time series analysis, etc.), BT-SFS is used with Multi-class SVM...

 

 

 

 


 

 

Back to Yan Ke's Homepage..