Enhanced Short-Term Voltage Stability Assessment with the Incorporation of Impedance Information Learning

Short-term voltage stability (SVS) assessment in practical power systems is typically implemented via threshold-based empirical criteria. However, such criteria usually require a relatively long observation time window of 1~10 s to reliably assess system stability status, which makes it difficult for them to meet the efficiency requirement of online monitoring. Data-driven solutions based upon advanced machine learning technologies provide a promising alternative to achieve much faster online SVS assessment. Following the general data-driven idea, this paper develops a data/knowledge jointly driven approach to improve the accuracy of SVS assessment without compromising the online assessment efficiency. An impedance-based criterion with physical significance is first introduced. This criterion enables the extraction of impedance information of system Thevenin equivalence and load buses, which is taken as part of physics-informed inputs of the whole assessment approach. Further, system-responsive time series data acquired from transient processes are taken as additional inputs. By feeding such combinatorial inputs into a convolutional neural network (CNN), a deep neural classifier is built to conduct online SVS assessment. Finally, numerical tests performed on both a simplified system and the Nordic test system confirm the efficacy of the proposed approach.