Online transient stability margin prediction of power systems with wind farms using ensemble regression trees

Abstract

Summary A new method for online evaluation of the transient stability of wind farms incorporated system based on random forest regression is proposed in this paper. The data before contingency was employed as the inputs instead of the post fault features. The critical clearing time is employed as the transient stability boundary, which determines how stable the system is after the given contingency. The mapping function between the pre‐contingency conditions and the corresponding critical clearing time is modeled as ensemble regression trees model, which consists of lots of base learner. Through the bootstrap method and the random selection of variables in the training process, the problem of dimensionality disaster can be avoided naturally without the need to specifically select features. The out‐of‐bag error generated during the bootstrap process is used for parameter selection and variable importance measures. Case study on the New England 39‐bus system incorporated wind farms and IEEE 118‐bus system shows that the proposed method has a strong prediction accuracy and generalization ability.

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

“…The method is tested on a well-known test system to simplify analysis and focus on the key interpretability theme. However, accurate estimation of transient stability status using larger systems has been proven in [38], [39]. For the proposed method to be implemented in larger systems, the additional computational burden primarily relates to the increased number of RMS-TDS required to construct ML models at each busbar and thus capture the entire stability boundary of a system.…”

Artificial Intelligence Techniques for Power System Transient Stability Assessment

“…Furthermore, the IEEE 68-bus system (16 machines) and IEEE 145-bus system (50 machines) ought to be mentioned here, as these are somewhat larger test case power systems that also serve as benchmarks [17]. However, these are far less popular among researchers, with certain notable exceptions, e.g., [18][19][20][21][22]. Finally, it can be stated that there are several other test case power systems that have been used for TSA and related analyses, but these are often not fully disclosed and almost always lack certain information.…”

“…Several papers proposed different kinds of autoencoders (stacked, denoising, variational), e.g., [8,16,29,30]. Mi et al in [22] proposed a special bootstrap method and the random selection of variables in the training process for tackling the curse of dimensionality. This is still an active area of research, with autoencoders leading the way forward.…”

“…Training RF involves a significant number of hyperparameters. Many approaches to power system TSA, involving decision trees or random forests, have been proposed, e.g., [22,[41][42][43][44][45].…”

“…Ensemble learning has been a very popular approach to power system TSA, e.g., see [17,28,40,46,[49][50][51][52]. Furthermore, several authors have used different tree-based ensembles [22,41,43], or special bagging ensembles [16,32,53]. This is still an active area of research.…”