CNN-Based Inversion Method for Saturation Current in Current Transformers Under DC Bias

In high-voltage direct-current (HVDC) transmission and large-scale power-system operation, DC-bias effects can drive current-transformer (CT) cores into premature saturation, distorting the secondary current and seriously jeopardizing the reliability of protective relaying and metering. To address the limited fitting capability and robustness of conventional compensation approaches in the presence of nonlinear distortion, this paper proposes a convolutional neural network (CNN)-based inversion method for CT saturation current. First, a simulation model is built on the PSCAD/EMTDC platform to generate a dataset of saturated, distorted currents covering DC components from −50 A to +50 A. Then, a CNN with a three-layer one-dimensional convolutional architecture is designed; leveraging local convolutions and parameter sharing, it extracts features from current sequences and reconstructs the true primary current. Simulation results show that the proposed method accurately recovers the primary-current waveform under mild-to-severe saturation, with errors within 2%, and exhibits strong adaptability and robustness with respect to both the polarity and magnitude of the DC component. These findings verify the effectiveness of CNNs for CT-saturation compensation.