الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 96 |  |  |
| | | from | 96 | | |
Abstract
The purpose of this thesis is the application of artificial intelligent classifiers to discriminate among the types of power transformers’ faults as follows:
The first part of the thesis describes dissolved gas analysis (DGA), a common method used to diagnose transformer faults. The DGA methods, such as IEC Code, Rogers’ ratios, Duval triangle, and key gas methods, failed to interpret the transformer faults in some cases and had poor diagnostic accuracy. Therefore, the researchers try to enhance diagnostic accuracy by combining the traditional DGA techniques with artificial intelligence and optimization techniques. Still, they also have a complex way of interpreting transformer faults. In the current work, a classification learner toolbox in MATLAB presented several Classifiers to classify the transformer faults and construct a classifier model used to diagnose some other test samples. The classification learner in MATLAB is so easy to understand and implement in classification applications. Several data transformations were carried out to investigate their effect on diagnostic accuracy to identify which transformation method can achieve the highest diagnostic accuracy. The results indicated that the ensemble bagged classifier with raw data (data without any transformation) had the highest diagnostic accuracy of the transformer faults, reaching 83.4 %.
Finally, malfunctions may occur due to electrical, thermal, or mechanical stresses on the transformer insulation system from insulating oil and paper. DGA considers a common method for diagnosing transformer faults. Although the IEC Code, Rogers Ratio, and Duval triangle are the traditional DGA methods, they develop poor diagnostic accuracies. Optimization methods are used to enhance the performance of the artificial intelligence of traditional DGA to improve the high accuracy of diagnosing power transformer faults. Still, it individually does not give high diagnostic accuracy. Therefore, a transformer fault diagnosis smart system (TFDSS) was developed in this work to increase the high analytical accuracy of recent DGA methods based on comparing the output of four DGA methods such as code tree 2020, modified IEC and Rogers’ ratio method, and Neural pattern recognition. The smart system developed a diagnostic accuracy (89.12%), higher than the highest diagnostic accuracy created by neural pattern recognition (86.01%).