LSTM-based prediction of fault event sequences

Due to the interaction between AC and DC systems, the transient characteristics of power systems have become increasingly intricate, leading to higher safety risks. Transient simulation that yields high-dimensional and sparse data, serves as the foundational approach for investigating system dynamics. However, there is still a lack of effective methods for mining knowledge of simulation data. To address the above limitations, this paper proposes a knowledge extraction and application method for massive simulation data. Initially, a general method is proposed for applying the event-driven architecture in transient simulation. On this basis, an AC line auto-tripping mechanism is designed to simulate the development of faults. Subsequently, characteristic events (CEs) are predefined and continuously recorded throughout the simulation so that information can be quickly extracted from simulation results. After obtaining extensive sequences, the paper utilizes natural language generation techniques to process the textual sequences. An LSTM-based model has been built to learn the logic of CEs and predict the CE at the next time point. Finally, the validity of the proposed method is verified using the case of the IEEE 39-bus system.