Research On Key Technologies Of Fault Diagnosis And Prognostic For Power Electronic Circuits

With the increasement of modern power electronic devices and their complex structures,and as the the important foundation of the smart gird,reliability of power electronic devices is crucial to the safe and stable operation of large-scale power grids.Research on fault diagnosis and prediction technology of power electronic circuits can greatly enrich the PHM theoretical system,reduce the losses and maintenance costs caused by circuit faults in engineering applications,and promote the development of intelligent maintenance systems for power electrical equipment.Based on the theory of compressed sensing,deep learning,and particle filtering theory,this paper takes advantage of the advantages of traditional methods,and studies fault feature extraction,intelligent classification methods and fault prediction algorithms in fault diagnosis and prediction of power electronic circuits.The main results of this paper are as follows.(1)Power electronic fault signal preprocessing method based on compressive sensing is proposed.In order to achieve efficient and rapid fault diagnosis,the amount of data of power electronic circuit signals should be reduced and valuable fault feature information should be retained.This paper studies the feasibility of extracting fault feature parameters of power electronic circuits is discussed in the compressed domain.It is clear that the compressive sensing theory is applicable to the compression and reconstruction of power electronic circuit signals.Through the calculation and analysis of different signal feature of the circuit signals,it is verified that the signal feature parameters in the compressed domain are highly related to the original signal feature parameters.The feature parameters in the compressed domain can replace the original signal feature parameters of the power electronic circuit to perform fault test and diagnosis.(2)Two multi-fault diagnosis methods suitable for power electronic circuits are proposed.The fault features of multi-soft and multi-hard faults in power electronics are relatively similar,which causes the difficulties to diagnose correctly.In order to improve the accuracy of fault diagnosis,fault feature extraction and fault classification methods are considered,and the two fault diagnosis methods: based on JADE-SAE and based on WPE-ELM are studied.In the fault diagnosis method based on the JADE-SAE,the circuit time domain fault feature parameters are optimized for the compressed circuit signal firstly,and the original high-dimensional time-domain fault feature parameters formed by multiple measurement points of the circuit are reduced by using the JADE algorithm to further reduce the dimension fusion.The joint depth features of power electronic circuits are explored using the stacked automatic encoder model,and Softmax models are used to complete the classification of the circuit's multi-soft failure modes.In the fault diagnosis method of power electronic circuits based on WPE-ELM,the compressed circuit signal is first extracted using wavelet packet decomposition to extract the energy spectrum feature vector of the measurement point signal,and the fault feature dimension is reduced through PCA.Finally,different fault modes are separated by ELM.Simulation and physics experiments show that the proposed method is rapid in diagnosis,less in parameter setting,and diagnostic performance is better than traditional BPNN,SVM and other intelligent diagnostic methods.(3)A health feature parameter extraction method based on equivalent analysis was proposed.In order to eliminate the coupling influence of operating conditions on characteristic parameters,the influence laws of operating conditions on effective fault characteristic parameters under the same working conditions were analyzed,and equivalent analysis methods were studied.Based on intelligent algorithms such as BPNN,the relationship between failure feature parameters under standard operating conditions and the actual operating conditions and fault feature parameters under actual operating conditions is established.The equivalent value of fault feature parameters under standard operating conditions is used as a circuit-level health feature parameter under variable operating conditions.The health parameters extracted are no longer affected by operating conditions and are only related to degradation of circuit performance and can reflect the health of the circuit.(4)A multi-order particle filter fault prediction method is proposed.The power electronic circuit has the characteristics of strong nonlinearity,high frequency and high noise,and the fault evolution behavior exhibits non-linearity,time-varying,uncertainty,etc.The traditional prediction algorithm is difficult to achieve high-precision prediction,and the particle filter algorithm has advantages to deal with these problems.Using historical time series data of health feature parameters,combined with system process noise,intelligent algorithms such as LSSVM are used to establish multi-order state space models based on circuit health feature parameters for particle filter algorithms which reflect the correlation between the current state and the previous steps.The multi-order particle filter algorithm can dynamically describe the performance degradation of the circuit and achieve accurate prediction of the future circuit state.The experimental results show that the new method has good prediction performance under different model parameter settings,and it is feasible and effective for fault prediction of power electronic circuits.