Interpretable Deep Learning-Enabled Dominant Instability Mode Identification in Sending-End Power Grids with Multiple HVDC Links

In high-renewable sending-end power grids with power delivered via multiple HVDC transmission lines, transient AC/DC system faults may introduce severe power interruption within the grids. This could lead to composite instability phenomena with complex couplings, involving voltage instability, rotor angle instability, and frequency instability. In this circumstance, if the dominant instability mode is not accurately identified, erroneous stability assessment results and ineffective stability enhancement countermeasures may be derived. To cope with this problem, this paper proposes an interpretable deep learning-based approach to intelligent dominant instability mode identification and efficient explanations of key control parameters affecting system stability. First of all, a convolutional neural network is introduced to classify dominant instability modes based on temporal feature learning under diverse transient scenarios. Then, the Shapley additive explanation method is introduced to quantify the influence of different control parameters of HVDC converter stations and renewable generation units on specific instability modes. Analysis of variance is further performed to verify the statistical significance of these parameters. Numerical simulations are conducted on the modified CEPRI-RAS test system, where the test results demonstrate the efficacy of the proposed approach.