Hydraulic Turbine State Prediction Method Based On Extreme Learning Machine And Transfer Learning

Due to the adjustment of China’s energy structure,the role of electric energy in our daily life is increasingly obvious.With the rapid development of such energy,the complexity of power grid operation increases with the increase of power generation,so the security and stability of power grid operation is crucial.The hydraulic turbine unit can effectively reduce the pressure of the power grid,alleviate the peak valley conflict and other problems that are easy to occur in the power grid,and improve the power quality.During the operation of the hydraulic turbine,the failure of a part of the unit will cause serious consequences.Therefore,it is essential to ensure the safety of water turbine units and timely conduct fault diagnosis and state prediction for water turbine units.This paper presents a new method for predicting the state of hydraulic turbines.The depth residual network and migration learning are mainly introduced to build the hydraulic turbine state prediction model proposed in this paper.At the same time,the cavity convolution module,jump connecting line and extended exponential linear unit activation function are used to improve the depth residual network to improve the state prediction performance.First of all,in order to preprocess the turbine signal,the parameters of the variational mode decomposition are optimized by the Drosophila optimization algorithm,the turbine signal is decomposed by the variational mode decomposition after parameter optimization,and the accuracy is verified by short-time Fourier.transform.Secondly,kurtosis criterion is introduced to reconstruct the turbine signal.In order to fully extract the features in the turbine signal,the reconstructed turbine signal is processed with continuous wavelet transform to obtain time-frequency diagram.Considering that histogram equalization can highlight image features and increase local contrast,this paper uses histogram equalization to enhance the features of time-frequency image obtained through continuous wavelet transform to obtain time-frequency image after feature enhancement.Finally,in order to realize the state prediction of hydraulic turbine,the convolution module,residual module and activation function of the depth residual network are improved.Considering the characteristics that migration learning can solve problems in different but related fields with existing knowledge,combined with migration learning,the state prediction model of hydraulic turbine proposed in this paper is obtained.The turbine state prediction model proposed in this paper is used for prediction,and then compared with several groups of comparative experiments to evaluate the performance of the turbine state prediction model proposed in this paper.