Minor Fault Detection For Sensors In AHU With Kernel Principal Component Correlation Analysis And Double Layer Bidirectional LSTM

The signals collected by the sensors directly affect the work of the air-conditioning system,which accuracy is the key to ensure the normal functioning of the air-conditioning system.The structure of the air-conditioning system is complex,the parameters are coupled with each other,and the sensors are prone to various faults during operation.Therefore,the fault detection and diagnosis(FDD)of the air-conditioning system sensors is great important.Although the existing FDD is quite mature,there has been little research on minor soft faults of sensors.If these faults are not discovered in time,they may affect the normal operation of air-conditioning systems,and even result in loss of energy and a quality reduction of the indoor environment.If these faults can be detected in time in the early stage,they may create conditions for subsequent diagnosis and repair,and avoid the damage to the air conditioning system from aggravating the faults.To accurately detect minor faults of sensors in air handling unit(AHU)sensors,a method for detecting minor faults of sensors is studied.Firstly,kernel principal component correlation analysis(KPCCA)is proposed to extract sensor faults characteristics to solve the problem of excessive redundancy in extracting sensor fault feature parameters of AHU.It is found that the kernel parameters can directly affect the performance of KPCA for the extraction of minor faults features by exploring the principle of KPCA.Therefore,particle swarm optimization algorithm based on the correlation method for kernel parameters is proposed with the principle of information entropy,and the optimized kernel parameters are used for KPCA to extract the relevant variables of the air-conditioning system,which is kernel principal component correlation analysis.Secondly,double layer bidirectional long short-term memory(DL-Bi LSTM)is proposed to solve the time-series dependence of minor sensor faults.The minor fault feature data extracted by the KPCCA is converted into a sequence with time function by using a sliding window.double layer bidirectional long short-term memory structure is proposed,incorporating the temporal correlation of faults.This structure includes twolayer forward and backpropagation double-loop network structure,which enhances robustness while ensuring the acquisition of bidirectional information.In addition,this structure specially sets dropout layers to prevent deep learning from overfitting.Finally,a novel fault detection method based on kernel principal component correlation analysis-double bidirectional long short-term memory(KPCCA-DL-Bi LSTM)is proposed to detect minor faults of AHU system sensors timely and accurately.The detection accuracy of KPCCA-DL-Bi LSTM is 30.66%,29.00%,28.00%,and 27.00%higher than that of KPCA for drift faults at the level of 5%,10%,15%,and 20%,respectively.The detection accuracy of KPCCA-DL-Bi LSTM is 16.66%,17.34%,16.67%,and 17.33% higher than that of standard LSTM for drift faults at the level of 5%,10%,15%,and 20%,respectively.The experimental results show that KPCCA-DLBi LSTM has better fault detection accuracy and stability for minor drift faults of the sensors.