Research On Transformer Fault Diagnosis Based On Rough Set And Random Forest Algorithm

The power industry regarded as the cornerstone of China’s rapid economic development,makes a significant role in the development of various fields.While,the transformer is the most necessary equipment of the power system.The normal working condition of transformer can directly determine the operating quality of the power system.Therefore,due to the dissolved gas analysis(DGA)technology in transformer oil,the accuracy of transformer fault diagnosis is getting particularly important.In view of the low accuracy of transformer fault diagnosis methods,the limitations of single intelligent algorithm and high date quality requirements,a transformer fault diagnosis model based on rough set and random forest algorithm is proposed in this paper.The Random Forest(RF)is a new machine learning method which combines the Integrated Learning and the Decision Tree.The RF is different from the neural network algorithm,the extreme learning machine,the classification regression tree,the support vector machine and other single algorithms.Random Forest as an integrated learning algorithm with better learning performance can overcome the limitations of each single algorithm.Mentioned in this paper,the integration algorithm and DGA technology are combined to realize the function of on-line monitoring transformer status and fault diagnosis,while,the developing defects can be found at the original stage of the fault,so that transformer faults can be more intelligent,faster and more accurate detection.so as to realize more intelligent,faster and more accurate detection of transformer fault.In order to further improve the diagnosis accuracy and speed according to the basis of the algorithm advantages.A new method combining random forest algorithm(RF)and rough setting algorithm is proposed in this paper.First of all,the fault sample data obtained by DGA method is combined with a variety of transformer gas data processing methods such as three ratio method and Rogers method to obtain multiple groups of gas ratio,improve the characteristic quantity of the data,expand the data attributes,and then get the preprocessing decision table data through normalization processing.Then,rough set algorithm is used to reduce the preprocessed data set to obtain the simplest decision table data,obtain the core attributes,reduce data redundancy and improve data quality.This not only ensures the high quality of data,but also avoids excessive redundancy of data,and can improve the accuracy of final fault diagnosis.Finally,in order to verify the superiority of the fault diagnosis model based on rough set and random forest algorithm proposed in this paper,the preprocessed decision table data and the simplest decision table data are substituted into the transformer fault diagnosis model of four algorithms:random forest(RF),AdaBoost,limit learning machine(ELM)and classified regression tree(CART),The diagnosis accuracy and diagnosis time under two categories and eight different diagnosis models are simulated respectively,and the conclusions are obtained by comparative analysis.Except the RF,the Classification Algorithm is also used and discussed in this paper.Three practical and representative machine learning algorithms for simulation comparison are also introduced.At the same time,the data sets before and after rough set processing are substituted into the four classification algorithms for simulation comparison.Based on the diagram analysis of eight groups of simulation results,it is concluded that the transformer fault diagnosis model based on rough set and random forest algorithm has the highest diagnostic accuracy of 99.76%,and the diagnosis speed is also advantageous,so the model is more efficient,more accurate,and has higher practical value.Figure[24]table[20]reference[80]...