Sparse dictionary learning for transient stability assessment

Abstract

Transient stability assessment (TSA) has always been a fundamental and challenging problem for ensuring the security and operation of power systems. With more power electronic interface resources integrated into the grid and large renewable energies, the stability of the power system is jeopardized. Therefore, TSA of the power system should be considered in advance to keep the system running stable. In recent years, with the development of artificial intelligence (AI) technologies such as artificial neural network (ANN), support vector machine (SVM), and Markov decision process, TSA has improved dramatically. In this study, a sparse dictionary learning approach is proposed to improve the precision of the classification accuracy of transient stability assessment in power systems. Case studies of TSA using multi-layer support vector machine (ML-SVM) and long short-term memory network–based recurrent neural network (LSTM-RNN) are discussed as benchmarks to validate the proposed method. The stable and unstable dictionary learnings are designed based on datasets obtained by simulating thousands of different time-domain simulation (TDS) scenarios performed on the New-England 39-bus system in the PSAT (power system analysis toolbox) toolbox. Stable and unstable dictionaries are developed based on the K-SVD approach. The testing dataset contains both stable and unstable samples which steps into the sparse coding process to obtain the indexes. Compared with the indexes, the system’s final TSA is targeted. The proposed method exhibits satisfactory classification accuracy in transient stability prediction and provides the ability to reduce false alarms both in positives and negatives of the power system.