Research And Application Of Forecasting Algorithm For Vibration Failure Of Hydraulic Turbine

With the improvement of water resources utilization rate,in our country,the scale and installed capacity of hydropower stations are also increasing,and the safety of hydropower units is becoming more and more important.The failure of the unit will cause harm to personnel and equipment,and even cause serious economic losses.Therefore,it is of great significance to carry out research on fault diagnosis and fault prediction of hydraulic turbines.At present,the related research in the field of fault diagnosis is mostly the current fault and status of the target equipment,while the prediction of future failures of the system is less researched.In terms of hydropower units,due to the lack of fault samples of the turbine,the fault prediction of the turbine is even less.There are few practical engineering needs that existing fault diagnosis algorithms can no longer meet.In order to maintain the unit in a targeted manner and improve the safety of equipment,it is urgent to research and develop fault prediction methods for hydraulic turbines.With the development of smart technologies such as big data cloud computing,fault prediction technology has great research significance.This is not only a manifestation of the gradual intelligence of hydropower units,but also the development direction of the entire country’s energy construction.Deep learning has attracted attention because of its powerful time series forecasting ability and its advantages in real-time processing of large data samples.Aiming at the problems of low accuracy and difficult prediction of vibration fault diagnosis of hydraulic turbine system,this paper proposes a hydraulic turbine system fault prediction method based on deep learning LSTM long short-term memory network combined with DBN deep confidence network,which combines wavelet packet energy band with time-frequency domain indicators Combining information,extracting high-dimensional fault statistical features,using the adaptive feature extraction capabilities of the DBN deep network to perform high-dimensional feature representations on the original fault data,accurately judging the types of faults,and predicting the future with the powerful predictive ability of LSTM on time series signals Possible vibration failure of the system.The specific content includes:(1)To clarify the research background and significance of this subject,and to introduce in detail the current fault prediction methods and the current research status of fault diagnosis for hydraulic turbines.(2)Analysis of failure mechanism of hydraulic turbine equipment.The basic structure,working principle and fault characteristics of hydraulic turbine equipment are introduced in detail,and the types of hydraulic turbine vibration faults are analyzed from the three factors of hydraulic,mechanical and electromagnetic.(3)Feature extraction and analysis of vibration fault signals.Time-frequency domain features are obtained by time-frequency domain analysis method,and based on wavelet analysis and wavelet packet decomposition theory,a three-layer decomposed wavelet packet energy spectrum is obtained,and a variety of features are combined as the feature vector input of the predictive network model later.(4)Establishment of hydraulic turbine fault prediction model based on LSTM-DBN.Introduced the structure and principle of deep learning long and short-term memory network and the optimization method of deep neural network;and established a fault prediction model from the aspects of data preprocessing,model parameter setting,prediction process and framework,etc.Paving the way for the verification of the following algorithm model.(5)Case simulation research and analysis.Carrying out engineering simulation experiments,and comparing and analyzing algorithms,showing the advantages of long and short-term memory networks and deep confidence networks,accumulating statistical data for predicting the form and degree of possible damage to unit equipment components,and verifying the effectiveness and engineering of the algorithm proposed in this paper.