Sequence Impedance Prediction for Grid-Connected Converters Based on Bi-LSTM

The large-scale application of voltage source converters (VSCs) to power grids may negatively impact their stability. Stability analysis methods based on impedance modeling offer a theoretical framework to analyze and address this issue. However, under conditions of uncertainty regarding the internal structure and parameters of VSCs, existing impedance measurement techniques using frequency scanning are limited in terms of data acquisition and accuracy. In practice, the power output demand for VSCs in grid-connected operation tends to vary, making it challenging for frequency scanning schemes to account for changes in operating conditions. To address this challenge, this paper proposes a sequence impedance prediction solution based on a bidirectional long short-term memory (Bi-LSTM) method, which requires only limited data to accomplish high-precision prediction of the overall impedance of VSCs under multiple operating points. Unlike the existing dq-based impedance prediction schemes, the sequence impedance method will greatly simplify the prediction process, and the physical meaning is clear. In addition, this paper applies trained model to a small-scale VSC model through transfer learning for experimental validation. The process saves computational resources while maintaining prediction accuracy and demonstrates the adaptability of the proposed model.