Post-fault Transient Stability Assessment Based on k-Nearest Neighbor Algorithm with Mahalanobis Distance

To enhance the security of power system, fast and accurate transient stability assessment methods are highly in demand. Many works have been done to explore the stability classification by searching and matching based on the post-fault PMU measurement data. Generally, the Euclidean distance is adopted to evaluate the similarity of different time series data. However, the correlation of the sample attributes is ignored in the classifier based on Euclidean distance. Besides, due to the imbalance feature of the transient stability assessment, costs of false classifications are unequal. For example, classifying the stable samples into unstable ones introduces less operational risks than classifying unstable samples into the stable. To handle above technique challenges, this paper proposes a k -Nearest Neighbor algorithm using Mahalanobis distance for transient stability assessments, in which the correlation of critical time series data is carefully considered. To strengthen the cost-sensitive feature of the proposed classifier, training processes are optimized regarding an area under the curve index. Finally, a case study is presented where the IEEE 10-machine 39-bus system is used for tests. Results of classifiers with different distance are compared. Test results validate the efficacy of the proposed k-Nearest Neighbor algorithm using Mahalanobis distance. Besides, it is also demonstrated that using Mahalanobis distance, the stability assessment can be accurate and robust even only based on a few measured states in form of time series data.