| With the development of the energy industry,electricity consumption continues to increase,and the actual power supply cannot meet the demand,accompanied by problems such as aging of substation equipment lines,thermal failures of equipment,and inability to update and replace equipment in a timely manner.Among them,the troubleshooting and diagnosis of thermal faults has gradually become one of the key issues in substation fault diagnosis.Based on the actual background,this article conducts in-depth research on fault diagnosis of substation equipment.Firstly,an in-depth analysis was conducted on the actual system network structure of the substation,and six subsystems were divided for analysis based on different system divisions in the network structure.Among them,analyze the causes of thermal faults in each system and address how to diagnose and prevent the occurrence of thermal faults.Based on previous research,this article adopts machine learning algorithms to address the current situation of problems that occur on site in substations.Secondly,on the basis of machine learning theory,in order to solve practical problems,the idea of Ensemble learning is adopted to avoid the existence of over fitting.By partitioning the sample dataset,different systems will use different sample data for pre processing operations such as feature extraction,Newton interpolation transformation,and normalization.Finally,the decision tree algorithm in machine learning is used as the basic learner,and the Ensemble learning idea is adopted.Finally,the Random forest algorithm is followed,and then the Bagging idea is used to combine the weak classifiers of multiple Random forest to form a strong classifier.This can not only improve the accuracy,but also avoid the problem of low credibility of the results caused by the over fitting of small sample data.Finally,through practical examples,taking the fault data of a certain substation in the north as an example,a specific analysis is conducted on faults such as transformers,transmission lines,circuit breakers,transformers,and reactors in the substation.The traditional gas content ratio method,gas three ratio,four ratio and Random forest optimized fault classifier are used for comparative analysis of the results,and it can be concluded that Ensemble learning Random forest method has more accurate diagnosis results.The diagnosis results of three methods: SVR,BP neural network and integrated Random forest method are compared,and it is known that the Ensemble learning method can effectively improve the recognition accuracy.Then,through the diagnosis and analysis of the transformer,transmission line,reactor,disconnector and circuit breaker faults using the methods proposed in this paper,the diagnosis results of each system can achieve good results.According to the Confusion matrix results output by Matlab,the recognition degree is more than 90%.The transmission line fault selection is processed by the method of Ensemble learning combined with wavelet basis SVM,and the recognition degree is more than 97%,which can prove that the idea of Ensemble learning is effective for substation fault diagnosis. |
