Research On Fault Diagnosis Of Hydropower Units Based On Symplectic Geometric Mode Decomposition And SOM Neural Network

As the core equipment of hydropower station,the healthy and stable operation of hydropower unit is related to the safety of hydropower station and even the whole power grid.About 80%of the faults of hydropower units can be reflected by vibration signals.However,due to the fact that vibration signals are usually affected by many factors,it is very difficult to extract signal features and identify faults.This paper proposes a fault diagnosis method for hydropower units based on Symplectic Geometry Mode Decomposition(SGMD)and SelfOrganizing Feature Maps(SOM).The main research contents are as follows:Firstly,the vibration signal of hydropower unit is decomposed by symplectic geometric mode decomposition algorithm.Symplectic geometric mode decomposition algorithm is an effective method to deal with complex signals based on symplectic geometric similarity transform.This method has good decomposition effect under the premise of ensuring the integrity of the essential characteristics of the original signal.On the basis of symplectic geometric mode decomposition,this paper introduces Pearson correlation coefficient and proposes an adaptive SGMD algorithm to solve the problem that component reorganization needs to set the threshold of periodic similarity artificially and cannot be reconstructed adaptively.The adaptive SGMD algorithm can not only separate the noise in the original signal without artificially setting parameters to achieve noise reduction,but also reconstruct the similar components with the same characteristics in the initial component to obtain the final decomposition result.Through simulation signal verification,the adaptive SGMD algorithm has good decomposition effect and noise robustness for nonlinear vibration signals.Compared with empirical mode decomposition(EMD)and variational mode decomposition(VMD),the adaptive SGMD algorithm has better decomposition effect and noise robustness.Finally,the fault diagnosis model of hydropower unit is studied.Based on the above signal feature extraction,Self-organizing Feature Maps(SOM)is used as a pattern recognition method.Because it has good learning ability and generalization ability,the algorithm is selected to send the feature vector to the SOM neural network for fault diagnosis.In order to verify the classification effect of SOM neural network,it is compared with support vector machine,back propagation neural network and variational prediction model.The results show that SOM neural network has better classification effect.Based on adaptive SGMD and SOM neural network algorithm,this paper proposes a fault diagnosis model for hydropower units,and verifies the effectiveness of the model through actual vibration data of hydropower units.Secondly,the bubble entropy is used to extract the features of the decomposition results of the adaptive SGMD algorithm.Bubble entropy is an improvement of permutation entropy,which reduces the dependence on parameter optimization in entropy calculation and has higher stability and discrimination.The symplectic geometric component obtained by adaptive SGMD decomposition is extracted by bubble entropy as the fault feature vector.The bubble entropy is compared with the feature extraction methods such as sample entropy,permutation entropy and fuzzy entropy.The results show that the bubble entropy has good feature extraction ability,and has better effect than the other three entropy.Thirdly,SOM neural network is used to identify the feature vectors obtained by the above method.The SOM neural network is different from the general neural network based on the reverse transmission of the loss function.It optimizes the parameters of the network by constantly adjusting the weights,and has good learning ability and generalization ability.The feature vector is sent to the SOM neural network for pattern recognition.Finally,this paper combines the adaptive SGMD algorithm,bubble entropy and SOM neural network to establish a fault diagnosis model for hydropower units.The validity of the model is verified by the actual vibration data of hydropower units,and it is compared with least squares support vector machine,back propagation neural network and variable prediction model.The results show that the model has the highest diagnostic accuracy for various types of faults and overall faults compared with the other three models.