Transient Stability Assessment Based On Voltage Time Series And Machine Learning

With the rapid development of China's economy,electric energy has become an indispensable part of various industries.The importance of smart and strong power grid is self-evident.As the power grid gradually develops toward the UHV AC-DC hybrid large power grid,the power grid structure becomes more and more complex,and the massive data of the power system brings new opportunities to the study of transient stability.How to make better use of data and build models becomes new challenges for the grid.Based on the above background,the main research contents of this thesis are as follows:1)The analysis of the relationship among the trajectories of voltage,electromagnetic power and rotor angle after disturbance is carried out.Since the voltage trajectory is closely related to the rotor angle trajectory,it can also be used to predict transient angle stability.And in the power system transient stability assessment based on data mining,it is difficult to accurately capture the trend of the electric quantity.This thesis directly uses the voltage time series data,instead of the snapshot data to form dataset,and train the classifier.2)In this thesis,the fuzzy C-means clustering algorithm is used to identify the typical bus voltage variation pattern after disturbance in stable and unstable conditions.The similarity of the voltage trajectory to the pattern is taken as features,and support vector machines is trained by the similarity value to predict steady state.To improve the ability of single weak classifier to fit data,it is proposed to use the adaboost method to form a strong classifier to improve the classification effect.The results of simulation test show that the proposed method is suitable for wind power grid-connected systems.The classification performance has been greatly improved after using adaboost.3)In view of the previous algorithm using simpler fuzzy cluster-ing for rough feature extraction,this thesis introduces a more elaborate shapelet method,which extracts the shapelet feature from the electrical quantity time series acquired by the PMU after the fault,and then formulates the decision tree model to assess whether the system is unstable.In view of its slow offline training,this thesis proposes to accelerate the extraction of shapelets using PSO.While ensuring high classification accuracy,it also provides a mechanistic explanation of the system instability.