Research On Condition Evaluation And Fault Prediction Diagnosis Technology Of Power Transformer

With the development of China’s economy,the scale of power grid is expanding and the number of power transformer is increasing.The safety of power grid is related to the national economy and people’s livelihood.Power transformer is one of the key equipment in power grid and it is of great significance for the safety of power network to ensure its safe and reliable operation.At the same time,the monitoring system of power transformer is constantly improved and the monitoring data is increasing.It is of important research value and significance to effectively use monitoring data to further improve the accuracy of power transformer status evaluation and fault prediction.Graph Neural Network is a special machine learning algorithm for graph data,which can fully mine the relationship between nodes and realize machine learning on graph.In this paper,the graph of the relationship between data is transformed into graph data,and then Graph Neural Network is used for relationship mining.Graph Neural Network and other neural networks are used to deeply excavate the monitoring information of transformers and further improve the state evaluation,dissolved gas concentration prediction and fault diagnosis technology of transformers,so as to improve the safe operation level of transformers.Firstly,this paper establishes a transformer condition evaluation system by selecting data from transformer electrical test data,dissolved gas data,insulation oil test data and operation data.The weights of the primary indexes are determined based on the analytic hierarchy process.Considering the possible relationship between the high-level indicators,this paper establishes the relationship diagram between the high-level indicators,further uses Graph Neural Network for deep mining,and then constructs the evaluation model.This model overcomes the problem that the analytic hierarchy process does not consider the correlation between indicators and improves the accuracy of evaluation.Next,the effects of LSTM,GRU and CNN neural networks on transformer oil-dissolved gas concentration prediction are compared in this paper.Then,the Pearson correlation coefficient between time series of gas concentration is used to construct the relationship diagram of gases.Graph Convolution Network is used to deeply mine the relationship between gases,and then a combination prediction model is constructed based on the time features extracted by LSTM and other neural networks.Through comparative experiments,the prediction error of the improved model on the test set is further reduced and the prediction effect is better.Finally,a transformer fault diagnosis model based on transformer characteristic gas diagram is established.From the point of view of graph classification,this model uses Graph Neural Network to discover the possible relationship between fault feature gases,and combines mult ilayer linear neural network to build a fault diagnosis model.Compared with the fault diagnosis model based on Back Propagation Neural Network,the accuracy of fault diagnosis is improved.