Research And Development Of Condition Early Warning System For Auxiliary Equipment Of Power Station Based On Deep Learning

The operation status of auxiliary machines in power stations directly affects the safety and economy of power production,and the auxiliary machine status warning can buy valuable time for operation and maintenance personnel to take measures to reduce fault losses and bring great economic benefits to power generation enterprises.Aiming at the problems of high coupling,difficulty in modeling and accurate early warning of the auxiliary system,this paper takes deep learning algorithm as the main technical breakthrough point,combines signal processing technology and data mining technology,designs and develops a set of early warning system of auxiliary equipment status of power station.Practical engineering application shows that this system can effectively monitor the state of auxiliary equipment,provide an early warning mechanism when the operating state of the unit is abnormal,and realize the safety monitoring of the main auxiliary equipment of the boiler.The research work and conclusions of the thesis mainly include:(1)For the common failures and failure mechanisms of key auxiliary equipment in power stations,the laboratory axial fan experiment platform was built to realize the simultaneous acquisition of multiple signals such as vibration,noise and current;the axial fan failure simulation experiments were designed and carried out to simulate four types of failures:rotor unbalance,rotor misalignment,rotor friction and duct damage,and the multi-operation data set of axial fan was formed;by observing the time domain waveform and spectrum distribution of the signals,the operating status of the fan can be quickly identified.(2)Research on vibration and noise signal noise reduction and feature extraction methods.Given the shortcomings of traditional time and frequency domain analysis,the wavelet threshold method was used to reduce the noise of the vibration and noise signals;a three-layer wavelet packet decomposition was performed on the vibration signal,and the normalized energy ratio of eight frequency bands were extracted as the state representation of the vibration signal;A noise signal feature extraction method based on 1/3 octave was proposed for noise signals,which makes up for the leakage of Fourier transform spectrum by windowing,and the extracted 1/3 octave sound pressure level spectrum was corrected by A-weight,and the results can more truly reflect the auditory characteristics of human ears.(3)An early warning method of auxiliary engine state based on stacked sparse denoising auto-encoder was proposed.The 1/3 octave characteristics of noise and energy characteristics of vibration wavelet packet under normal working conditions are used as model inputs to train and optimize the SSDAE network.A quantitative characterization method of auxiliary engine health based on similarity algorithm and probability fusion algorithm of D-S evidence theory is proposed,and an early warning strategy is designed.And the samples of different fault conditions were input into the model for verification.The results showed that the model can effectively distinguish between normal state and fault state,and can give early warning before the state deteriorates,which has engineering application value.(4)The functional and architectural design of the power station auxiliary engine status warning system was completed,and the software functional modules were developed based on LabVIEW and Python,and the installation of the system was finally realized.The actual engineering application cases show that the developed system can meet the basic requirements of power station auxiliary engine status monitoring and warning,and has certain practicality and advancement.