Research On The Feature Selection And Machine Learning Algorithm For Fault Diagnosis Of Air Conditioning

The use of machine learning algorithms to detection and diagnose heating ventilation air conditioning systems can detect and diagnose air conditioning faults in a timely and effective manner,ensuring efficient operation of the system and reducing energy consumption.However,the air-conditioning system is complex and there are many kinds of faults.The machine learning algorithms still needs improvement in the applicability of air conditioning fault diagnosis.For example,how to determine the possible faults in the air-conditioning system in a timely manner,how to broaden the scope of application of the diagnostic method when not easily obtaining the marker sample,and how to combine the experimental state with practical problems in the engineering design.In response to the above issues,the content of this paper is as follows:(1)A variety of feature selection methods are used to characterize air conditioning fault data.In view of the shortcomings in the current chiller fault diagnosis process,different feature selection methods are used to characterize the chiller fault data,and a support vector machine is used to complete the classification.Through comparative analysis,important feature subsets in the chiller fault diagnosis are obtained.The final result can more effectively troubleshoot the chiller.(2)By extending the sequence forward selection algorithm,this paper proposes a fault detection method combining cost-sensitive and sequence forward selection algorithm,and uses backward tracing sequence forward feature selection algorithm to select out-of-sequence and consider cost-sensitive features.The experimental results show that the classification accuracy of the experimental algorithm is higher than that of the equivalent literature.At the same time,this work achieves an effective combination of experimental design and practical engineering problems,which fills in the gap between theoretical air conditioning fault detection and diagnosis methods and real-world air conditioning fault detection and diagnosis applications.(3)Aiming at the condition of few samples with marked faults in the training set,an air conditioning fault detection and diagnosis method based on semi-supervised learning is proposed.Using a small number of labeled samples to train the support vector machine,the samples with high class-label confidence in each category were selected from the unlabeled samples are added to the training sample set,and the information of the unsupported samples in favor of the support vector machine was used to improve the performance of the classifier.The results show that the hybrid algorithm can achieve semi-supervised learning and achieve more than 90% fault diagnosis rate under the condition that there are few fault marker samples.