Power dispatching fault set automatic generation based on Attention-Bi-LSTM and XGBoost

Power system faults bring serious threatens and challenges to the safe and stable operation of the system. The traditional methods of generating predictive fault sets mainly rely on manual experience and the generated fault sets are often mismatched with the system safety characteristics. Motivated by this issue, this paper proposes an automatic fault set generation method based on Attention-Bi-LSTM and XGBoost. Firstly, a system operation risk model is established to calculate power equipment fault severity and fault risk. Then utilize the Attention mechanism to enhance Bi-directional Long Short-Term Memory (Bi-LSTM) model and combining it with eXtreme Gradient Boosting (XGBoost) model to construct a probabilistic combined prediction model for the risk of power dispatching faults with weights determined by the error reciprocal method. Finally high-risk fault sets are then automatically generated based on the obtained predictions in terms of risk probability and severity magnitude. The proposed method is verified in a practical power system in China and the results show that the fault sets generated can better reflect the actual system operating state, improve the fault prediction efficiency and provide support for the safe and stable operation of the power system.