Study On D-S Evidence Theory And Its Application In The Transformer Fault Diagnosis

Transformers are widely used in power grids,and real-time sensing of their operating status is one of the key steps to eliminate potential faults.Due to the diversity of transformer manufacturing processes and operating environments,which are prone to different kinds of faults,electricity enterprises have been spending a lot of manpower and material resources in transformer maintenance.For the uncertainty fault problem of transformers,a transformer fault diagnosis model based on D-S evidence theory is established in this dissertation.First,the related information fusion techniques based on D-S evidence theory are deeply analyzed and studied.Secondly,it is clarified that resolving conflicting evidence and constructing the basic probability assignment function are the main tasks.Finally,using the improved D-S evidence theory,a transformer fault diagnosis model is established to provide a theoretical basis for improving the accuracy of transformer fault diagnosis.The main research contents of this dissertation are as follows:（1）The dissertation proposes a new method of evidence synthesis,mainly from the perspective of modifying evidence sources,to address the phenomenon that conflicting evidence will cause the synthesis results to be counter-intuitive in evidence synthesis.The correlation matrix is constructed by first measuring the correlation degree between the evidence using the Spearman rank correlation coefficient;then the original basic probability assignment function is modified by introducing correction coefficients;finally the modified multiple sets of evidence are fused using the traditional D-S evidence theory combination rules.The numerical examples show that the proposed method has good adaptability in dealing with highly conflicting evidence synthesis and has higher target recognition rate,which is very important for the performance improvement of the information fusion system.（2）From the perspective of fuzzy science,a deeper investigation is conducted on how to assign BPA reasonably.A generalized trapezoidal fuzzy number-based BPA assignment method is proposed,and the principle is to calculate the corresponding affiliation degree based on the constructed model to complete the assignment of BPA.The classical Iris dataset in the UCI database is selected as the object of study,and the specific process of the method is introduced in detail.The basic target recognition experiments and the recognition experiments in the closed world with and without noise environment are carried out respectively.Through multiple sets of comparison experiments,it is proved that the method in this dissertation has excellent properties in dealing with small-scale data sets and can be widely applied in practical engineering.Compared with other methods,this method can accurately obtain quantitative indexes,properly solve the problem of subjectivity in the evaluation system,meet the decision making needs of decision makers,and ensure the scientific nature of the applied method.（3）From the perspective of information fusion,the improved D-S evidence theory proposed in this dissertation is applied to transformer fault diagnosis.By analyzing the relevant characteristic quantity parameters in transformer faults,the basic probability assignment function based on fuzzy affiliation is constructed for feature-level fusion,and the experimental results verify that the proposed transformer fault diagnosis model based on evidence theory is feasible and effectively guarantees the power supply safety and power quality of the distribution system.