Ensemble learning based multi-fault diagnosis of air conditioning system

The failure of air conditioning systems is random and uncertain, with one or more faults occurring simultaneously at any given time. Factors such as difficulty in collecting fault data and the singularity of existing diagnostic models all impact diagnostic performance. Ensemble learning based fault diagnosis strategy for an air conditioning system was proposed. By using generative adversarial networks (GANs) to enrich the training database and combining Bagging and Boosting to reduce model variance and bias, the final output was derived by integrating the diagnostic results of multiple models. Additionally, Spearman correlation analysis and feature importance ranking were utilized to identify the features that have the greatest impact on the model so as to further enhancing its performance. Experimental results demonstrated that this method can effectively diagnose single and multiple faults even in the absence of sufficient training data. An overall accuracy of 98.54 % was achieved with a Hamming loss as low as 0.0047.