Transient Stability Awareness Of Power System Based On Long Short-term Memory Network And Scale Invariant Feature Transformation Algorithm

Power system is an important energy security system to guarantee national economic development and social stability.The analysis and research on its operation security and reliability has been a hot issue that many relevant scholars are competing to study.In recent years,along with the large range of the interconnection project overall deployment and the sustainable development of electricity market,the operation of the grid environment increasingly complex,the power system security and stability margin is more and more small,all areas of power system network topology and mathematical model are close to or even already in pathology,brought the safe and stable operation of power system more severe challenges.It is of great significance to judge the transient stability of power system quickly and accurately for preventing the stability damage of power system effectively caused by large-scale disturbance and even further causing the breakdown of power system.Therefore,it is urgent to develop an effective tool which can quickly and accurately judge the transient stability of the system.The traditional method of power system transient stability judgment is based on the extended equal area criterion,which quantifies the stability margin of the measured curve of the system obtained by the system characteristic measuring device.The effectiveness of such traditional methods largely depends on the accuracy of the rapid prediction of the disturbed characteristic trajectory of the system network after large disturbance.However,with the increasing size of the power network topology,it is difficult for this type of method to be applied to the power network with multi-source heterogeneous characteristics today.And rising in recent years the development of machine learning and deep learning methods relying on its excellent learning performance,has been active in the areas of power system analysis,this method can avoid power network topology complex problems,power grid dynamic data as input,set up power system transient analysis problem of the system state vector and accurate mapping relations between the network status,for power system transient stability analysis provides a new train of thought.Aiming at many problems in the transient stability analysis of power system,the following work is done in this paper:(1)Based on the similarity of spatio-temporal motion law of voltage phase trajectory,a traditional clustering algorithm based on space is proposed to perform clustering analysis on voltage phase trajectory,and the identification of system generator consimilarity can be accomplished quickly and accurately according to the clustering results.(2)A personalized long short-term memory network peer-to-peer system based on dynamic time warping and long short-term memory network is proposed for accurate prediction,which effectively improves the accuracy and timeliness of trajectory prediction.(3)From a data point of view,by analyzing the correlation between the source-network features,the scale-invariant feature transformation algorithm is used to construct a suitable index to quantify the boundary features of the transient stability of the system.