Blocking method with PSO-SVDD for differential protection of power transformer

Multiple features fusion through artificial intelligence holds the potential to enhance the reliability of transformer protection. However, the existing methods encounter challenges due to the data scarcity of inrush current and internal fault. To address this, the paper proposes a blocking method of non-fault scenarios for differential protection, employing particle swarm optimization-support vector data description (PSO-SVDD). The approach utilizes several geometric features extracted from the differential current-excitation voltage curve (DEC) of normal operation to fully characterize similarities among non-fault scenarios, such as inrush current, and CT saturation caused by external fault. By exclusively utilizing normal operation data to represent all non-fault scenarios, the challenge posed by the scarcity of non-fault data is overcome. PSO-SVDD, selected as a one- class classification algorithm, is trained using geometric features from normal operation data, thereby avoiding the limited availability of fault data. The region inside the SVDD hypersphere serves as the blocking domain of differential protection. Additionally, a fine-tuning method of the feature boundary is introduced to enhance the robustness of the blocking domain by the small sample of scarce scenarios. Before the model is applied to unseen transformers, offline detection using its normal operation data is conducted to evaluate the performance of the blocking domain in. In case of misidentification, the misidentified sample is utilized as the support vector for fine-tuning the feature boundary. In practical application, if iron core saturation or fault scenarios are misidentified, the feature boundary is fine-tuned offline using the misidentified sample with the similar method above. PSCAD simulations and dynamic simulation experiments validate the superior performance of the proposed protection method through the comparison with several existing methods.