Measurement error evaluation method for voltage transformers in distribution networks based on self-attention and graph convolutional networks

Accurately evaluating the error of voltage transformers in distribution networks is crucial for the safe operation of power systems and the fairness of electricity trade. This paper uses the connection relationship between distribution transformers and voltage transformers to predict the secondary voltage of voltage transformers through the secondary voltage of transformers, constructing a voltage transfer characteristic model between the two to achieve accurate evaluation of voltage transformer errors. To address the challenge of extracting complex nonlinear features from multivariate electrical data, a combined model of a self-attention mechanism and a graph convolutional network (GCN) is proposed. The self-attention mechanism captures global dependencies among power parameters, while the GCN effectively constructs the multivariate data structures in distribution networks. By integrating both approaches, the model can fully extract the intrinsic features of the data as well as the hidden dependency information between data points. Additionally, to prevent gradient vanishing as the combined model’s structure deepens, a multi-head residual structure is introduced to enhance the self-attention mechanism. Experimental results show that compared to a single model, the proposed combined model reduces the mean squared error by 82.35% and increases the coefficient of determination R² by 9.07%, demonstrating significant accuracy advantages in voltage transformer error evaluation.