Fault Diagnosis Of Hydroelectric Sets Based On Singular Value Decomposition And Deep Belief Network

As a representative of clean energy,hydropower resources,its reliable supply can effectively solve the environmental pollution problem and maintain economic and social sustainable development.Therefore,it is particularly important to ensure the safe and stable operation of hydroelectric sets in actual production.According to statistics,more than 80%of the faults in the hydroelectric sets are reflected in the vibration signal.Therefore,effectively analyzing and extracting the vibration signal characteristics of the hydroelectric sets and accurately diagnosing the fault are great significance to the safe and stable operation of the hydroelectric sets.This paper takes the vibration fault of hydroelectric sets as the research object,proposes a fault diagnosis method for hydroelectric sets based on singular value decomposition and deep belief network multi-classifier.The main research contents and results are:Firstly,for the problem of low signal-to-noise ratio of early fault signal of hydroelectric sets,a denoising method based on singular value decomposition is used to reduce the noise in the vibration signals of hydroelectric sets and verified by simulation.Construct the Hankel matrix of the vibration signal containing noise and decompose it into singular values.In the phase space reconstruction process,the four threshold selection methods for the effective singular values of the constructed Hankel matrix are studied.The noise reduction effect of the differential spectrum method is better than other three singular value threshold selection methods.At the same time,the noise reduction effect of the singular value differential spectrum method is compared with the noise reduction effect of the wavelet transform,and it is concluded that the noise reduction effect of the singular value differential spectrum method is significantly better than the wavelet transform.Secondly,for the problem of the complex construction of the vibration signal feature vector of the hydroelectric sets,a method of feature extraction of the hydroelectric sets based on singular value decomposition is adopted and verified by simulation.The attractor orbit matrix is constructed using the noise-reduced vibration signal,and its singular value decomposition is performed,and the decomposed singular value sequence is used as the eigenvector of the vibration-reduced noise signal.Through simulation analysis,it is concluded that this method can reliably and effectively obtain the true characteristics of various states.Thirdly,for the problem that traditional shallow neural network classifiers are prone to overfitting and local optimization during classification,a fault diagnosis method for hydroelectric sets based on deep belief network multi-classifier is adopted.Studied the structure and training process of the deep belief network.The setting of related parameters such as the number of nodes,weights,and offsets of the deep belief network in the process of establishing the fault diagnosis model of hydroelectric sets is discussed.The network diagnosis process of the deep belief network is also discussed.Finally,an engineering example is used to verify the fault diagnosis method of hydroelectric sets proposed in this paper.This method is compared with classifiers such as deep belief network sub-classifier,BP neural network,multi-class support vector machine,and singular value median method,wavelet transform and other noise reduction methods,and verified with samples of unknown fault types.All the results show that the fault diagnosis method of hydroelectric sets proposed in this paper has higher recognition accuracy and faster calculation speed,can identify the fault type of hydroelectric sets more stably,reliably and efficiently,and it has some application value in actual production.