Data Domain Adaptation for Voltage Stability Evaluation Considering Topology Changes
Haosen Yang, Ziqiang Wang, Robert C. Qiu
Abstract
Recently, intensive efforts have been poured into the real-time voltage stability assessment (VSA) by machine learning methods using measurement data. However, one serious and open problem of learning methods for VSA is that they heavily suffer from sudden topology changes and significant variation of system parameters. If a sudden topology change occurs, traditional learning methods usually need a retraining process that consumes excessive time and large amount of pre-labelled post-change data. To address this problem, this paper proposes an online adaptive VSA method based on data domain adaptation, which can rapidly adapt to the new topology after the change by limited amount of unlabelled post-change data. Besides, to improve the VSA accuracy, this method presents an enhanced temporal convolution network (ETCN), which can both extract the long-term time series characteristics of VSA and the key information at current time, as the primary neural network framework. The proposed method performs high accuracy and efficiency in both stages before and after topology changes, as well as the weak requirement for the amount of post-change data and high robustness for measurement noise. Case studies in different testing systems and the comparison with other learning algorithms demonstrate the effectiveness and advantages of the proposed approach.