Research On Fault Early Warning Method For Two Power Equipments Based On Relevance Vector Machine

The safety and reliability of power equipment is an important prerequisite to ensure the safe and stable operation of power system,it is of great significance to strengthen the monitoring of the operation status and realize the early fault warning and diagnosis of power equipment.Different from traditional diagnosis methods,data-driven fault early warning technology can discover the early signs of equipment failure to prevent the occurrence of power equipment failures by digging out the underlying laws of the data,learning from the samples,and establishing a model to predict the relationship between the data.Thus,the operation state of power equipment is predicted based on this model,the early signs of power equipment failure can be found as early as possible,and then the early warning signal can be given to prevent the occurrence of power equipment failure.In order to further improve the accuracy of the power equipment fault early warning method,the relevance vector machine(RVM)algorithm is introduced into the power equipment fault early warning.First,this paper introduces the RVM algorithm and analyzes its performance,which shows that the algorithm can solve nonlinear system problems well and has good sparsity.At the same time,the power system data characteristics and data processing methods are analyzed to provide a basis for fault feature extraction of early warning models.Then,two models used in fault early warning that match the actual situation are established based on the data characteristics of different power equipments.Model 1 predicts normal operating parameters and gives the fault early warning based on the differences between the actual and predicted values.Model 2 provides fault early warning after predicting the fault trend.Finally,this paper verifies the application effect of the proposed warning model based on the sampling data obtained from the supervisory control and data acquisition(SCADA)system and the chromatographic data of transformer oil.According to the characteristics of the SCADA data of wind turbines,an early warning model for overheating failure of the main shaft bearing based on RVM was proposed.The model uses the RVM regression algorithm to predict the normal temperature of the main shaft bearing and calculate the Mahalanobis distance according to the actual operating temperature.If there is a potential fault in the main shaft bearing,the Mahalanobis distance obtained by the model will continuously exceed the alarm value,so as to provide early warning of potential faults.Moreover,the application of RVM in transformer fault early warning was studied based on the data of dissolved gas in transformer oil.That is,the RVM regression algorithm was used to predict the dissolved gas content in the oil,and the multi-kernel learning RVM algorithm was used to identify the status of the transformer,further analyze the examples whose results verify that the model can effectively predict the gas content in the oil and correctly diagnose transformer faults.