Research On Fault Diagnosis And Prediction Method Based On Knowledge Graph Of Transformer

As the most important transformer equipment in the conventional island of nuclear power plants,the operating state of the transformer directly affects the safety and stability of the nuclear power plant.Nuclear power plants at home and abroad have repeatedly experienced shutdown accidents due to transformer failures.The main transformer of nuclear power plants has also been included in the key supervision scope of the National Nuclear Safety Administration.At the same time,in accordance with the intelligent development of nuclear power plants and the requirements of Industry 5.0 construction,the maintenance of transformers is also urgently changing to the state maintenance and intelligent maintenance.Carrying out research on transformer fault diagnosis,fault prediction and expert decision,timely discovering latent faults of transformers,and accurately locating the root cause of the faults are of great significance to the promotion of intelligent maintenance of nuclear power plants and the guarantee of intrinsic safety.At present,there are many researches on transformer fault diagnosis and prediction algorithms,but they often focus on model selection and parameter optimization.They do not consider the problems of small sample size,unbalanced samples,and difficulty in obtaining related parameters in actual projects.At the same time,explosive growth Most of the transformer operation and monitoring data are stored in unstructured and semi-structured text.Due to the complexity of the text structure and the lack of effective knowledge organization methods,a large amount of valuable fault information is difficult to mine;at the same time,traditional technology is difficult to mass Heterogeneous data and fault diagnosis and prediction methods are systematically integrated.Based on an in-depth analysis of the status of nuclear power plant transformer fault maintenance,this paper conducts research on related algorithms and data for transformer fault diagnosis and prediction.For structured data,based on oil chromatographic data,the transformer fault diagnosis and prediction related models were studied;for semi-structured data,the transformer fault text classification model was studied;finally,based on the knowledge graph technology architecture,fusion of transformer multi-source heterogeneity Data and transformer fault diagnosis,prediction and text classification models,research on transformer fault diagnosis and prediction methods based on knowledge maps,to a certain extent,it solves the small sample size,uneven distribution,and difficulty in obtaining related information in the actual operation of transformers.Problems such as the unavailability of semi-structured data and the imperfect knowledge base of fault experts.The main research contents of this article are as follows:(1)In order to improve the accuracy of the fault diagnosis model under small-scale and unbalanced samples,a double support vector machine fault diagnosis model based on chemical reaction optimization is studied.The model incorporates a variety of dissolved gas fault diagnosis feature parameters in oil as input Parameters to ensure the comprehensiveness of fault feature coverage;the use of restricted Boltzmann machine for data preprocessing to ensure the effective identification of feature parameters,the use of binary tree multi-classification dual support vector machine as the classifier,the fault prior knowledge and machine The combination of learning methods improves the classification ability of the model under unbalanced samples,and uses the optimization method combining chemical reaction optimization and cross-validation methods to improve the generalization ability and robustness of the model.Finally,the effectiveness of the model is verified by actual failure sample cases and random tests.(2)In order to meet the higher safety requirements of nuclear power plants,the fault prediction is used as the starting point.For the problem that the associated indicators in the prediction sample are difficult to obtain and the scale is limited,the study is based on phase space reconstruction and weighted least squares support vectors.The prediction model of dissolved gas in the transformer oil of the engine,the model uses weighted least squares support vector machine as the classifier,and the phase space reconstruction method based on chaos theory is used to preprocess the input parameters,mining the inherent laws and characteristics contained in the historical data To improve the prediction performance of the prediction model when the sample size is small and the associated parameters are difficult to obtain;the optimization method that combines chemical reaction optimization and cross-validation is used to ensure the generalization ability and robustness of the model.Finally,the comparison test under different models and samples shows that the model can better analyze the development trend of the future state of the transformer in the case of small sample size and limited correlation indicators and meet the higher safety requirements of nuclear power.(3)In order to solve the problem that the point prediction cannot characterize the uncertainty of the results,based on the interval prediction theory and the point prediction model,the Bootstrap-based transformer oil dissolved gas interval prediction model is studied,and the weighted minimum based on the phase space reconstruction.The point prediction model of the square support vector machine is combined to construct a prediction model of dissolved gas in transformer oil that can be used for both point prediction and interval prediction.The model takes into account data noise and model errors,that is,it can describe the predicted Accuracy can also describe the uncertainty of prediction.Finally,through the actual case and the comparison test of different models,the performance superiority of the model is verified.(4)In view of the problem that the rich fault information in the semi-structured text cannot be obtained and used,a fault text classification model based on Word2vec and TF-IDF weighted text vectors is studied.The model uses a combination of hidden Markov word segmentation and dictionary word segmentation.Fault word segmentation in a way to improve the accuracy of text segmentation in professional fields;the text is converted into a feature vector of distributed representation through the Word2Vec method,taking into account the semantic relationship between vocabulary;TF-IDF is used to make the word vector output by Word2vec important Degree-weighted processing improves the accuracy of fault classification;adopts dual support vector machines as classifiers to classify faults in the list of accidents;and finally considers the characteristics of text multi-classification,sets up multi-class evaluation indicators on the basis of two-class indicators,and passes actual cases The classification performance of the model is verified.(5)In order to solve the problem of difficult integration and application of multi-source heterogeneous data and algorithms of transformers,the transformer fault diagnosis and fault prediction method based on knowledge graph is studied.By introducing the knowledge graph technology framework into the field of nuclear power transformer maintenance,from fault data acquisition,Fault knowledge extraction,fault knowledge fusion,fault knowledge reasoning research on transformer fault knowledge graph construction.Integrate various types of transformer fault parameters such as semi-structured data and structured data to improve the comprehensiveness of fault diagnosis and prediction data sources;BiLSTM-CRF named entity recognition method and TWSVM fault classification based on TF-IDF weighted text vector The combination of methods ensures the comprehensiveness of fault entity extraction;the XMLtoOWL method is used to realize the knowledge fusion of relational database and knowledge graph;the logic rule method based on a priori knowledge and the diagnosis and prediction method based on machine learning are combined to achieve transformer fault Comprehensive diagnosis and prediction;finally,the effectiveness of the method was verified by actual engineering cases.