Review of Neural Network-Based Transient Stability Assessment

With the increasing share of renewable energy generation in recent years, the inherent uncertainty associated with renewables has posed unprecedented challenges to power systems. Consequently, the demand for fast and accurate transient stability assessment (TSA) methods has become more critical. Neural networks, a significant area of inquiry within artificial intelligence, have been utilized in TSA challenges. This work provides a thorough review and comparison of three prominent TSA methodologies: time-domain simulation, transient energy function techniques, and artificial intelligence-based approaches. Representative studies examining diverse artificial neural network topologies, input feature selection methodologies, and hybrid modeling techniques are presented. At present, TSA of power systems faces numerous challenges: Although artificial neural network (ANN)-based methods have high accuracy, they still have incorrect predictions and lack constraints of physical laws; traditional time-domain simulation methods are computationally time-consuming, and transient energy function methods are difficult to apply to complex systems; meanwhile, the industry lacks standardized datasets and evaluation benchmarks, hindering performance comparison among models. The future development directions include adopting deep learning methods to improve the accuracy and speed of real-time assessment and promoting physics-informed neural networks (PINNs) to integrate physical laws into training for enhancing prediction accuracy.