8th International Conference on Advanced Power System Automation and Protection (APAP2019)

637 / 2019-04-12 15:32:11

A Transient Stability Assessment Method Using LSTM Network with Attention Mechanism

transient stability assessment; time series variable; LSTM; attention mechanism

the operation environment of modern power system is complicated, which puts high requirements for transient stability assessment. Recently, the machine learning method based on data analysis has been widely studied. However, most of the research is insufficient in processing the information of features and underutilized in the data mining capacity of learning machine. In this paper, the time series variables (the trajectory of bus voltage amplitude, power angle, rotor speed, electromagnetic power, and the unbalance of electromagnetic power and mechanical power) during post-fault are used as the inputs of the learning machine. The LSTM (Long Short-Term Memory) deep learning network with attention mechanism is applied to process the time series variables, to make full use of the prior information in the big data and overcome the “dimension disaster” problem of high dimensional data. The simulation results of 10 machine 39 bus system show that the proposed transient stability assessment method can process the time series variables effectively and improve the prediction accuracy.