Power Grid Transient Stability Margin Assessment Based On Long Short Term Memory Networks

Transient stability assessment is the main content of modern power system security analysis.With the rapid development of China’s economy,people need more and more power.The complex and changeable power grid operation conditions bring great challenges to the dispatcher to quickly and accurately evaluate the transient stability of power system.Traditional transient stability assessment methods,such as simulation time domain method,extended equal area method and so on,rely too much on the mathematical model of power grid,which makes the timeliness and accuracy of the analysis process difficult to match the development speed of power grid.With the wide application of wide area measurement system(WAMS),data support is provided for stability assessment of power system.Artificial intelligence method based on computer technology,with its strong learning ability and extremely short assessment time,provides a powerful tool for data-driven transient stability assessment of power system.At present,the artificial intelligence method generally uses the track information of the disturbed power grid for fast transient stability assessment,but uses relatively less steady-state information before fault.Therefore,this paper proposes coherent generators based on long short-term memory network(LSTM)by using the characteristic information of the whole stage time series.1)By collecting the whole stage time series characteristics as the input of the model,the time series evolution characteristics of power system faults are fully extracted;combined with the trajectory analysis method,the transient stability index of power grid is constructed to realize the evaluation of transient stability margin.2)Under the given expected fault set,LSTM is constructed to describe the mapping relationship between the grid characteristic quantity and the generator stability index,so as to realize the fast prediction of the generator stability index,and to gain time and provide decision support for the grid operation dispatcher.3)This paper uses LSTM to predict the voltage phasor trajectory information in the initial stage of power grid fault in ultra real time,and then extracts the trajectory features through the constructed trajectory offset feature plane,and uses DBSCAN density clustering method to cluster the predicted trajectory,and finally verifies the coherency identification results through extended equal area criterion(EEAC).The simulation results of IEEE-39 bus system verify the effectiveness of the above method.The proposed method can achieve fast prediction of generator stability index and accurate identification of coherent generator group through LSTM.Compared with other machine learning algorithms,it has better generalization ability,and can provide important reference for power grid operation dispatcher to take control measures.