Identifying the Core Aspect Ratios of Three-Limb Core-Type Transformers Using Machine Learning

Abstract

This paper proposes an approach to identifying the core aspect ratios of three-limb core-type power transformers. Analyzing transformer transients, such as inrush currents, is essential for ensuring uninterrupted power delivery, particularly in grids dominated by Inverter-Based Resources (IBRs). Topological-based transformer models, such as the Unified Magnetic Equivalent Circuit (UMEC) used in commercial simulators (e.g., PSCAD and RSCAD/RTDS), accurately represent transformers. However, these models require data on transformer core dimensions, which are often unavailable or proprietary. Consequently, despite the accuracy of topological-based models, they are less applicable in grid analysis. The proposed approach utilizes a Machine Learning (ML) algorithm, namely Extreme Gradient Boosting (XGBoost), to identify the transformer core aspect ratios using only current and voltage measurements when the transformer is in the steady state during no-load conditions. The performance of the proposed method is evaluated using MATLAB/Simscape for modeling transformers and MATLAB for running the XGBoost algorithm. It is inferred from the test results that the proposed method is able to identify the core aspect ratios of three-limb core-type transformers with satisfactory precision. The proposed method will enhance the accuracy of transient analysis, and it could be extended to cover four-limb and five-limb core-type transformers in the future.

Keywords

• Extreme gradient boosting (XGBoost)
• parameter identification
• transformer core aspect ratios
• unified magnetic equivalent circuit (UMEC)