Intelligent DGA-based Technique for Fault Identification of Power Transformers

Dissolved gas analysis is widely used for diagnosing faults in power transformers and has been extensively adopted by researchers to develop efficient fault diagnosis systems. In this regard, this paper proposes a novel DGA-based technique for identifying faults in power transformers. Using a dataset of 501 samples and covering six types of faults, an advanced Decision Tree-based classification method was applied to DGA data. To improve classification accuracy and reduce training time, an enhanced binary Crow Search Algorithm was employed for feature selection. Chaos mapping was integrated into the optimization process to avoid local optima. Additionally, the cross-validation technique was employed to ensure the randomized selection of data for training and testing the enhanced algorithm. The proposed method achieved a fault identification accuracy exceeding 92% for power transformers, outperforming most recently published fault diagnosis methods.