Research On Oil Temperature Prediction And Oil Temperature Anomaly Diagnosis Algorithm Based On Concept Drift

Transformer is one of the core equipment in power transmission system.It is very important to maintain the normal operation of transformer to ensure the normal transmission of power resources.The high oil temperature on the top layer of transformer is one of the main factors that cause transformer failure.Traditional transformer oil temperature fault diagnosis technology includes hot circuit model,oil color spectrum analysis and so on.However,due to the high cost of equipment,complex technical operation,complex parameter setting and other reasons,it is difficult to carry out large-scale promotion.In recent years,using machine learning to predict the oil temperature of oil immersed transformer,and then based on the algorithm to predict the value of oil temperature to diagnose the abnormal fault of oil temperature of transformer,is the hot direction of machine learning in the application of power grid.However,due to the concept drift in the oil immersed transformer data,the concept drift in the oil temperature data will make the algorithm predict the oil temperature error larger,and then make the reliability of the algorithm to diagnose the oil temperature abnormal fault reduced.In order to solve the above problems,this paper proposes two transformer oil temperature prediction algorithms based on concept drift.Algorithm 1:Based on the fixed window of transformer oil temperature prediction algorithm.Algorithm 2:transformer oil temperature prediction algorithm based on dynamic window mechanism.The above two algorithms can adapt.to the concept drift of oil temperature data,which makes the algorithm have good oil temperature prediction performance.On this basis,an oil temperature abnormal fault diagnosis method is proposed to realize the early warning of potential oil temperature abnormal fault defects of transformer and the alarm of oil temperature abnormal fault.The related contents are summarized as follows:Oil temperature prediction algorithm 1:its main idea is:1.In the process of oil temperature prediction,if the performance of oil temperature prediction of the basic learning device of the algorithm is poor,then use the latest oil temperature data to update the basic learning device,so that the updated basic learning device can adapt to the concept drift in the oil temperature working condition data.2.The weight update and competition strategy are used to optimize the performance of the algorithm.Oil temperature prediction algorithm 2:its main ideas are:1.Set two decision thresholds ? and ?,of which 1>?>?>0.In the prediction process,if the prediction accuracy of the basic learner is lower than the threshold ?,the model training set of the basic learner will be updated;if the accuracy of the oil temperature prediction of the basic learner is lower than the threshold ?,the corresponding model training set will be used to update the basic learner,so that the algorithm can adapt to the concept drift in the oil temperature data.2.The weight update and competition strategy are used to optimize the performance of the algorithm.Oil temperature abnormal fault diagnosis method:because the oil temperature data T1 predicted by the oil temperature algorithm is closely related to the oil temperature T2 of the system sensor.Based on T1,T2.a set of logic rules for transformer oil temperature fault diagnosis is summarized.This rule realizes the early warning of transformer oil temperature abnormal fault,and can analyze and diagnose different types of transformer oil temperature abnormal fault at the same time,which has good popularization.