A New Fault Diagnosis Approach for the Air Handling Unit Under the Imbalanced Data Environment

Most of the existing fault diagnosis methods for the air handling unit (AHU) do not consider the data imbalance problem arising in the AHU's actual operation, which leads to the unsatisfactory fault diagnosis accuracy. In order to solve the issue of the AHU's imbalanced fault datasets, this paper proposes a novel extreme learning machine (ELM) based fault diagnosis approach by infusing the SMOTE oversampling and the Bagging based ensemble learning. Specifically, the SMOTE oversampling is adopted to enrich the minority fault datasets to rebalance the fault dataset. On this basis, the ELM combined with Bagging based ensemble learning is further developed on the rebalanced fault datasets to improve the AHU's fault diagnosis effect. Multiple basic ELM diagnosis submodel are first established and the submodels with good diagnosis performance are then selected. The nearest neighbor subsets are constructed according to the fault samples to further define the weight factors, which are utilized to fuse the selected basic ELM diagnosis submodels, in order to build the final ensemble ELM diagnosis model. The detailed experimental results on the experimental dataset RP-1312 of the ASHRAE demonstrate the effectiveness of the proposed fault diagnosis method under the AHU's imbalanced data environment.