Research On Fault Classification And State Trend Prediction Of Hydraulic Turbine Based On Extreme Learning Machine

Hydraulic turbine unit is the core equipment of hydropower generation.Fault diagnosis and state detection of hydraulic turbine unit fs of great significance for the development of hydropower technology and the maintenance of the safety and stability of hydropower.This paper will focus on the common faults,feature extraction,signal analysis,fault diagnosis,state prediction and other issues of hydraulic turbine,taking empirical mode decomposition as the breakthrough point,combined with the extreme learning machine prediction method to carry out the fault classification and state prediction of hydraulic turbine.Firstly,the fault types and mechanism of hydraulic turbine unit are introduced.The main faults of hydraulic turbine are abnormal temperature,fault of oil slinger,blocking of grid connection,rotor grounding fault,mechanical fault and so on.This paper summarizes the mechanism,advantages and disadvantages of fault diagnosis methods such as fault tree analysis,fuzzy clustering,rough set theory and neural network method,and states time-frequency signal processing methods such as short-time Fourier transform,wavelet transform and empirical mode decomposition,which provide theoretical basis for the follow-up research.Then,EMD is used to decompose the signal of hydraulic turbine unit.According to the value of component characteristic parameters and the corresponding fault evaluation index,the component characteristic parameters with higher sensitivity are selected as the characteritics that can best reflect the operation state of hydraulic turbine unit,so as to lay a foundation for accurately judging the operation state of hydraulic turbine unit.A fault siganl separation method based on fast independent component is proposed.The simulation mixed data is separated by using fast independent component analysis.The feasibility of signal separation is verified and the basis for subsequent feature extraction is provided.Then,the ensemble empirical mode decomposition is used to extract the fault features,and the corresponding suppression methods for the end effect and mode aliasing that may occur in the process of fault extraction are developed.The original data sequence is extended into a ring data to suppress the end effect,and ensemble empirical mode decomposition is used to suppress mode aliasing.The efectiveness of this method in vibration signal feature extraction is verified by simulation.Finally,the fault diagnosis and state prediction model of hydraulic turbine unit based on ensemble empirical mode decomposition and extreme learning machine is proposed.The fault vibration signal of the components is decomposed into five intrinsic mode components and one residual component.The key components are selected and the corresponding eigenvalue values are calculated.Through the analysis of the corresponding characteristic parameters,the fault classification and operation state trend prediction of the hydraulic turbine unit are carried out.Finally,from the perspective of theory and practical engineering application,the effectiveness of the application of extreme learning machine model in fault classification and state prediction of hydraulic turbine is verified through comparative simulation analysis.