Research On Vibration Signal Noise Reduction And Trend Prediction Method Of Pumped Storage Unit

As the core equipment of the pumped storage power station,pumped storage units play important roles in the safe and stable operations.The load regulation of pumped storage units is frequent,and a variety of operating conditions alternate with each other,which is easy for the vibration of the units to intensify and occur frequently,and increases the risk of vibration accidents greatly,even leads to serious consequences such as flooding the station.Therefore,it is urgent to carry out on-line monitoring and trend prediction research on vibration status of pumped storage units.In this paper,the vibration signal during the operation of the pumped storage unit is taken as the research object,focusing on the scientific problems such as noise reduction of the pumped storage unit,nonlinear trend prediction and other scientific issues,with variational mode decomposition,BA algorithm,wavelet analysis and using neural networks as the theoretical basis,the defects and deficiencies of existing methods in theory or application are analyzed,and corresponding improvements are made.At the same time,the research results are applied to engineering practice,and a B/S architecture is designed and developed.The real-time vibration monitoring and trend prediction system of pumped storage units provides new ideas for the improvement of the state maintenance strategy of pumped storage units,and has certain theoretical and engineering significance.The main contents and achievements of the research are as follows:(1)In order to solve the problem of low data storage efficiency in vibration state monitoring,a distributed time-effect database for vibration state monitoring of pumped storage units was designed.The memory database Redis and the relational database My SQL are combined to quickly read and write real-time data and cache,and store historical data for long-term stable storage,thereby achieving efficient data storage and providing stability for subsequent prediction and analysis of unit vibration trends,Reliable pumped storage vibration data source.(2)In order to effectively extract the vibration signal characterizing the actual operating state of the pumped storage unit under the background of strong noise,a noise reduction method combining BA-VMD and wavelet threshold is proposed.Firstly,the BA algorithm is used to globally optimize the parameters K and α in the VMD decomposition,and the optimal [K,α] combination is obtained by adaptive search;then the optimized VMD method is used to decompose the original vibration signal into a series of IMF components;The autocorrelation energy function criterion selects the high-frequency IMF component dominated by noise and performs wavelet threshold denoising;finally,the low-frequency IMF component without wavelet threshold denoising and the high-frequency IMF component undergoing wavelet threshold denoising reconstruction is performed to obtain the vibration signal after noise reduction.The proposed method is applied to the noise reduction experiment of unit simulation vibration signals,and compared with a single VMD noise reduction and a single wavelet threshold noise reduction method.The experimental results show that the proposed method can effectively reduce the noise of the unit vibration signal.(3)In order to make up for the lack on the decision-making of pumped storage power stations afterwards and find out the signs of unit failures in time,this paper proposed a prediction model based on the mixed state trend of VMD and CNN-GRU,combining the processing methods of VMD signals with the prediction methods of the neutral network,aiming at the difficulties that the pumped storage unit vibrations appear to be non-stationary and are difficult to predict and analyze.Firstly,VMD is used to decompose the original nonstationary signal into more stable modal components,and then CNN-GRU hybrid neural network model is used to train each component,which reduces the complexity while improving the prediction accuracy.The proposed hybrid model was successfully applied to the vibration state trend prediction of a unit,and ablation experiments were conducted with CNN-RNN model,single CNN model and single GRU model to verify the effectiveness of the proposed hybrid model,which is the vibration trend of pumped storage unit Research on prediction methods has expanded new ideas.(4)Based on the above research results,the characteristics of B/S network architecture and front-end and back-end separation technologies were studied,and advanced application software systems with real-time vibration data monitoring,trend prediction and other services were designed and developed to verify the vibration signal drop of the pumped storage power station proposed in this paper The effectiveness of noise and trend prediction methods,some of the theoretical results have been integrated into the domestic large-scale pumped storage power plant unit vibration status monitoring and fault diagnosis system,which has effectively improved the operation and maintenance level of the power plant and effectively promoted the pumped storage power plant in the context of new energy.Intelligent development.