Intelligent Fault Diagnosis And Vibration Trend Prediction Of Hydropower Units

With the increasing proportion of intermittent new energy such as wind power and light energy connected to the grid,China’s energy supply structure has undergone significant changes.Under the multi-energy complementing of power grid,the dispatching of hydropower station gradually changes from the energy supply demand to the energy regulation mode,forcing the hydropower unit to expand the load operation range to meet the requirements of peak and frequency regulation compensation for intermittent energy supply.However,longterm and wide-ranging operation leads to poor operational stability of hydropower units,increasing the probability of faults occurring in these units.It makes the types of faults more complex and makes it difficult to capture information about unit faults.The traditional maintenance strategy of combining post-maintenance and scheduled maintenance can no longer meet the requirements for efficient and safe operation of the units.Conducting research on condition maintenance with fault diagnosis and condition trend prediction as the core is of great significance.It involves proposing a fault diagnosis system suitable for different key parts of a hydropower station and a high-precision condition prediction model.This approach enables the transition from passive response to early failure to active prevention of failure,ensuring the safe operation of the unit and the stability of the power grid.Thus,this paper focuses on several scientific problems related to the maintenance of hydroelectric units,including feature extraction of key components,identification of axis locus,and prediction of vibration trends.By conducting an in-depth exploration of the limitations of existing methods and models,the paper aims to enhance the condition maintenance system of hydroelectric units.Firstly,multiscale entropy theory and machine learning are used to construct fault diagnosis models suitable for different key components of hydropower units.Secondly,a new method for identifying unit axis locus is developed based on deep learning and transfer learning technology,using multi-source information fusion.Thirdly,a multi-step prediction framework for the vibration trend of hydropower units is proposed,based on a signal decomposition algorithm,rolling prediction strategy,and machine learning.The main innovations and research contents of this paper are as follows:(1)Due to the challenge of extracting fault features from vibration signals of hydroelectric units that are subject to the coupling effect of multi-source excitation,an improved nonlinear dynamic technique of measuring the time sequence signals complexity named improved multiscale attention entropy was proposed in the paper,which was based on entropy model theory and scaling technique.The technique was used to extract the fault features of vibration signals of hydroelectric units.On the basis of this technique,a new fault diagnosis model of hydropower units based on vibration signals is constructed by integrating improved multiscale attention entropy,T-distributed stochastic neighbor embedding and random forest algorithm.At the same time,the proposed diagnosis model is applied to the fault diagnosis cases of hydraulic and mechanical key components in the hydropower unit,which verifies the validity and generalization of the proposed model,and provides a new method and theory for the fault diagnosis of key components in the hydropower unit.(2)The axis orbit is a direct manifestation of the running state of hydropower units,and has become an important indicator in the running state analysis and fault diagnosis system of hydropower units.In order to identify the axis orbit accurately,a method of identifying the axis orbit of hydropower units based on image and signal features is proposed in this paper.Firstly,an image recognition method based on transfer learning and Xception is proposed to extract the image features of the axis orbit.Secondly,based on the two-dimensional displacement signal of the axis orbit,the signal features of the rotor axis locus were extracted by improved the multiscale attention entropy.Then,the signal feature and image feature are fused to establish the fusion mechanism of the image and signal features of the hydropower unit axis orbit.Finally,relying on the powerful recognition performance of random forest model,this paper realizes the efficient identification of unit axis orbit.The proposed recognition algorithm is applied to the case of sliding bearing fault,the effectiveness of the proposed method is verified,and the development direction of the identification of the hydropower unit axis orbit is expanded.(3)Aiming at the inability of traditional statistical models to accurately predict the vibration trend of hydropower units,this paper proposes a hybrid prediction model based on variational mode decomposition and stochastic configuration network.Firstly,the vibration data of the unit is decomposed into modal components of different frequencies by means of variational mode decomposition to effectively alleviate the instability of vibration trend.Secondly,the stochastic configuration network is used to predict different modal components,and the predicted values of all modal components are superimposed to form the single step prediction results of the unit.Finally,according to the recursion strategy,a multi-step prediction model of vibration trend of hydropower unit is built by adding the prediction results to the form of new input variables.In this paper,the proposed model is applied to the trend prediction cases of different types of signals such as pendulum signal and vibration signal of actual power plant units,and the results show that the proposed multi-step prediction model can accurately predict the vibration trend of power plant units and has great practical value for maintaining the safety of power plant units.