Data creation techniques and data-oriented models for dynamic security assessment of power system

Mahesh Pal Singh,Nidul Sinha

DOI: https://doi.org/10.1007/s00202-024-02319-6

IF: 1.63

2024-03-22

Electrical Engineering

Abstract:Advances in data-oriented artificial intelligence approaches open up the power system dynamic security assessment using machine learning (ML) models. The performance of the ML models depends on robustness (redundant and highly informative) of databases to handle security concerns under abnormal conditions and system constraints. Considering the current data availability and the need for more comprehensive methods to create dataset inclusive of all dynamically critical points of a power system, a maiden approach is proposed in this paper using intelligent and efficient Latin hypercube sampling technique. It employs a convex relaxation technique-based infeasibility certificate and examines highly informative regions for a specific secured operation threshold to construct a comprehensive database that encompasses all dynamically critical operating stages of a power system in a very efficient manner (off-line) that can be used for training ML models for real-time security assessment. The performance of the proposed technique is compared with Brute-force (BF) technique for IEEE 14 bus system only as BF is computational inefficient and intractable for larger systems and found to be very efficient in terms of data size, computational time, and efforts. The datasets created using the proposed technique for three standard IEEE systems (14, 57, 162 bus) are used for training seven ML models to determine if the system's current operating state is secure or not, including the margin of stability. Out of seven models, decision tree and random forest are found superior to others specifically for larger and complex systems.

engineering, electrical & electronic