Identification of Important Locational, Physical and Economic Dimensions in Power System Transient Stability Margin Estimation

Abstract

Increasing renewable generation can lead to significant spatial and temporal changes to the rotor angle stability boundary, such that critical contingencies may drastically change. Additionally, the inherent variability of renewables increases the number of operational scenarios that require stability assessment. This paper presents a methodology whereby a series of location-specific Decision Tree Regressors are trained, using power system variables to estimate the Critical Clearing Time (CCT) on a locational basis throughout a network. Permutation feature importance is used to reveal the most important power system variables for CCT estimation at each location (capturing aspects related to physical system characteristics, operational parameters as well as economic dispatch). Consequently, estimation of the duration and location of the critical fault can also be made -along with identification of important system variables that explicitly impact the critical fault. Results on the IEEE 39-bus network show accurate estimation of locational CCTs, with a mean absolute percentage error of 1.19% on average. Moreover, the mean absolute percentage error for the minimum CCT is 0.49%. An analysis of important power system variables is provided, demonstrating how the method can assist in the design of targeted locational interventions to improve the stability margin at specific locations.

Using SHAP Values and Machine Learning to Understand Trends in the Transient Stability Limit

“…In [25], authors propose the use of PFI [24] as the IML technique to extract understanding of the transient stability limit. PFI is limited in that the output is a ranked list of features based on the mean importance and the decrease in model performance when a feature is permutated.…”

“…A global explanation of the power system variables that impact the estimation of the critical fault (CCTmin) is provided using PFI (following the method presented in [25]). A similar analysis is conducted for the remaining 25 ML models.…”

“…Permutation Feature Importance (PFI) is a global technique [24], used in [25] to interpret DT models trained to predict the transient stability limit. PFI can provide feature importance, but not feature effects [21] and is limited in that feature importance is based on the decrease in model performance (i.e., is linked to the error of the model).…”

Power System Transient Stability Preventive Control via Aptenodytes Forsteri Optimization with an Improved Transient Stability Assessment Model

“…A TSA model can be built by machine learning algorithms, where the operational features of the system are taken as the input, and the assessment of transient stability is taken as the output. In [10], a series of location-specific decision tree (DT) regressors are trained for TSA, which leads to an improvement in the stability margin at specific locations. An imbalanced correction method based on a support vector machine (SVM) is proposed in [11], and this method is used to establish the TSA model.…”