Study On Abnormal Early Warning Method Of Hydraulic Turbine Based On Data Mining

With the improvement of the importance of clean energy in the energy structure,the hydropower project in China has been increasing.The stable operation of turbine units is of great significance for avoiding hidden dangers and improving the economic benefits of power generation enterprises.At present,hydropower plants have made some achievements in abnormal diagnosis and online detection,but the system can only provide intuitive monitoring data and some diagnostic methods.In this paper,under the background of more mature data mining technology and continuous accumulation of hydraulic turbine data,the research of water turbine abnormal early warning based on data mining is carried out.In the face of complex and changeable operation mode of hydraulic turbines,the time series clustering algorithm based on statistical characteristics is used to identify effective operation modes of hydraulic turbines.In view of the complex interaction relationship between the measuring points,the selection of the feature subset for the Gradient Boosting Decision Tree is introduced to the target prediction parameter,which not only strengthens the interpretability of the model,but also provides convenience for the later anomaly analysis.Based on the steady state data of the hydraulic turbine,the long and short term memory network(Long Short-Term Memory)in depth learning is introduced,and the multivariate time series prediction model of the measuring point of the hydraulic turbine is established by using the related measurement points as the characteristics.In the process of modeling,we combine all parameters to complete the optimization of the model.Then,based on the prediction model,we put forward the equipment state assessment and abnormal early warning mechanism,and set up a complete equipment abnormal early warning algorithm evaluation system.Finally,the effectiveness and feasibility of the algorithm is verified by taking the abnormal warning of fixed vane of a turbine as an example.Experiments show that the model can predict the abnormality in advance and avoid failure and unnecessary outage of equipment.The abnormal early warning algorithm system has great application prospects,and has important guiding significance for turbine abnormal prediction and fault diagnosis.