An online monitoring method for current transformers based on a gated attention Transformer model

With the continuous increase in the demand for power grid services, higher requirements arise for the online monitoring and condition assessment of current transformers as important equipment in the power grid system. Traditional electromagnetic current transformers require power-off for offline calibration, making it difficult to reflect their actual operating status, increasing operational complexity and costs, and affecting the accuracy of energy metering and the stability of the power grid. Addressing the issue of the lack of a standard transformer as a reference for online error monitoring of current transformers, this paper proposes an online monitoring method based on the gated attention Transformer (Gatedformer) model. This method uses multiple current data inputs, leverages the Gatedformer model to learn data features, and accurately predicts future standard values of current transformers. Specifically, the method embeds time point variables through the attention mechanism to capture multivariable correlations, uses a gated attention network to capture the long-term dependency characteristics of time series, and applies a feedforward network to learn nonlinear representations. The experimental results show that under the condition of predicting the standard current value window for the next day, the model achieves an average prediction error of 0.09% in the online monitoring of current transformers on the three-channel current dataset, outperforming existing models.