Interpretable Neighborhood Deep Models for Online Total Transfer Capability Evaluation of Power Systems

Abstract

Deep learning is considered to be a promising method for total transfer capability (TTC) evaluation due to its high speed and accuracy for nonlinear and complicated problems. However, the operating scenarios of power systems are complex and ever changing. It is difficult to train a single deep model using an overly complex dataset containing operating scenarios with high variability. In addition, making the deep model transparent to validate for operators is another challenge to its application in the power system. To enable the practical application of deep models in TTC evaluation, the operating sample space is clustered into simpler spaces via a two-stage clustering method, and the idea of neighborhood model for the deep TTC evaluation network (DTEN) is proposed. Then, the local post hoc interpretability for the DTEN based on first-order control variable sensitivity is presented. During the interpretability analysis, the quasi-steady state is introduced to assess the coupling effects among different inputs. Case studies on the IEEE 39-bus system and a real-world regional system in China validate the effectiveness of the proposed method.