Fault Diagnosis Of Hydropower Units Based On Multiscale Entropy And EXtreme Gradient Boosting

Under the continuous promotion of China’s "dual carbon" strategy,energy technology is constantly innovating,promoting the transformation of the new energy system,and also bringing new challenges to the safety of the power grid.As an important regulator of the new power system,the safe operation of hydropower is crucial,and once the unit fails,it will bring serious economic losses and casualties.The use of intelligent fault diagnosis technology to process the unit vibration signal can provide data support for monitoring and judging the health condition of the unit,and timely find and eliminate safety hazards,which has important engineering practical significance.This paper focuses on the problems involved in signal noise reduction,feature extraction and fault type identification in the fault diagnosis of hydropower units,and the main contents of the work are as follows:(1)Aiming at the problem that the effective information in the vibration signal of the hydropower unit operating environment is drowned by a large amount of noise,an adaptive noise reduction method of the vibration signal of the hydropower unit combining signal decomposition algorithm and entropy value theory is proposed.The method decomposes the signal by complete ensemble empirical mode decomposition with adaptive noise(CEEMDAN)and separates the noise dominant component from the signal dominant component by using the property of permutation entropy(PE)to reflect the degree of signal confusion;then the noise part of the noise dominant component is filtered by singular value decomposition(SVD)combined with the adaptive fixed-order method based on singular entropy;then all the components are fused and reconstructed to realize the effective noise reduction of the unit signal.Finally,after the experiments with the simulated signal and the unit vibration signal,the noise reduction capability and engineering practicality of the proposed method are verified.(2)Considering the nonlinearity,non-smoothness and sudden shocks of the hydropower unit vibration signal make its fault characteristics difficult to be recognized directly,the multiscale permutation entropy(MPE)and extreme gradient boosting(XGBoost)based fault diagnosis model for hydropower units is proposed,and the dung beetle optimizer(DBO)algorithm is introduced to optimize the model parameter selection in view of the large number of parameters and the combination of parameters that determines the model performance.Firstly,the skewness of the corresponding MPE is calculated by signal type as the fitness function of the DBO algorithm,and the parameters of the MPE are determined after comprehensive comparison;then the diagnostic error rate of the XGBoost classifier is used as the fitness function to determine its main parameters,which completes the optimization of the model parameters;finally,the bearing fault data set of Case Western Reserve University is used for experiments,and the DBO parameters are verified through the comparison of different models.The performance improvement of MPE-XGBoost diagnostic model after optimization was verified by comparing different models.(3)A new feature extraction tool,improved multiscale fractional-order weighted permutation entropy(IMFWPE),is proposed to improve the MPE,which has low sensitivity to structurally similar signals due to the lack of amplitude information,and its stability is reduced by traditional coarse granulation.The tool is improved with the coarse-grained method,and its good anti-interference and feature extraction ability is verified by the simulation signal and rotor fault data experiments.The IMFWPE-XGBoost fault diagnosis model is constructed experimentally using a rotor test bench to simulate the fault situation of hydropower units,combined with XGBoost classifier,and the results are compared with other models to prove that the proposed method performs best in fault identification.Finally,experiments in a real unit environment verified that it also has good fault diagnosis capability in engineering practice.