Research On Health State Prediction Of Steam Turbine Unit

As the core equipment of thermal power station and nuclear power station,it is very important to ensure the normal operation of steam turbine unit.If the health state of the turbine unit can be predicted,it will be beneficial for the crew to conduct timely control and reasonable maintenance,and reduce the probability of the turbine unit being in abnormal or serious fault state.This paper carries out research on the prediction of the health state of the turbine unit and establishes the prediction model of the health state of the turbine unit.The main work is as follows:(1)Pretreatment processing of data for steam turbine units.In order to realize health state prediction,it is necessary to preprocess data.The main work includes data cleaning,denoising,conversion and information dimension reduction.Firstly,the method of manually replacing outliers and missing values is used to clean the data,and the wavelet soft threshold denoising method is used to remove the interference noise of the cleaned samples.Then the Z-Score standardized operation was used for data conversion,and finally the correlation variance contribution rate fusion method was used to achieve data dimensionality reduction.After preprocessing,the data dimension is reduced,the characteristic information is retained,the proportion of effective information is improved,and the data requirements of subsequent work are met.(2)A combined prediction model has been proposed for data prediction of steam turbine units.To solve the problem of inaccurate prediction of a single model,Complete EEMD with Adaptive Noise(CEEMDAN)method,Long Short-Term Memory network(Long short-term Memory,Combined prediction model of LSTM and Support Vector Machine(SVM).The method of CEEMDAN decomposition of original data solves the problem that disordered data is not easy to predict and obtains stable and easy to learn decomposition quantity.After the decomposition quantity is predicted by LSTM and SVM,the weight of reconstructed data is determined by Lagrange multiplier method,which reduces the data reconstruction error.The results of CEEMDAN-LSTM-SVM model have higher accuracy after comparison experiment.(3)An improved condition discriminative model is proposed for the condition discrimination of steam turbine units.In order to accurately discriminate the state of the turbine unit,a Convolutional Neural Networks(CNN)based model is proposed in this paper.The feature extraction ability of the model is enhanced by adding a Gate Recurrent Unit(GRU),and an improved activation function PRe LU is chosen to reduce the overfitting problem of the model.The Beluga whale optimization(BWO)is applied to achieve fast optimization of the model parameters.Experimental verification and comparative analysis show that the state discrimination model has high accuracy.(4)Established a health status prediction model for steam turbine units.Based on the research of CEEMDAN-LSTM-SVM data prediction model and improved CNN state discrimination model,a steam turbine unit health state prediction model was built.After the preprocessing data is input into the model,the state label of the prediction time is output to realize the prediction of the health state of the steam turbine unit.Through the comparison experiment,the model evaluation index is introduced for comprehensive evaluation,and the results show that the state prediction accuracy of this model is higher than that of the unit.The comparison experiment is carried out,and the model evaluation index is introduced for comprehensive evaluation.The results show that the state prediction accuracy of this model is higher than that of the unimproved model.