Stability Assessment Machine Learning Models Interpretation Based on Surrogate Models and Clustering

Machine Learning has been wildly employed to power system stability assessment researches. However, most of machine learning models are lack of interpretability, which hampers their application in engineering. To overcome this deficiency, a model agnostic interpretation method is proposed for stability assessment machine learning. First, in the neighborhood of single data, a local surrogate model construction method is introduced to reveal the inner logic between the system state variables and stability level. Then, in order to get more general interpretations, a state data clustering method based on the Gaussian mixture model is proposed. The cluster interpretation is obtained by surrogate models of representative data. The interpretation results provide a reference for trusting the model and stability control. Finally, the interpretation method is applied to the voltage stability margin assessment machine learning model. The effectiveness of the proposed methods is verified.