Online Prediction Of Power System Transient Stability Based On Post-Fault Voltages Trajectory

With the gradual formation of interconnected power grid of our country, the requirements of security and stability to power grid are more stringent and it is significant to looking for a method that could assess the transient stability of the system quickly and accurately. With the development of artificial intelligence,machine learning has been a promising method to assess the transient stability of the system. The point of machine learning consider the transient stability test as a pattern recognition problem and hold the idea that there exist some mapping between the transient stability and the characteristics that describe operating status of the system. It can be obtained by machine learning. WAMS which is based on synchronized phasor measurements and modern communications technology monitors and analysis the vast power system. We can take the machine learning with the information contained in the data saved by the monitoring system and then find the methods to assess the transient stability.Firstly, this article discusses the application of intelligent learning on the assessment to transient stability of power system and analysis the advantages of the supporting vector machine on the objects of this research.It also explains the reason of the choosing the voltage track is the power angle values of it as electromagnetic transient is quicker than it as electromechanical transient and it simulates the voltage amplitude can predict the transient stability of power system more accurately in a shorter time than power angle.This article studies a lot of issues of online identification to transient stability of power system,like feature selection, combination forecasting model, learning online and assessment,combined of the intelligent learning to support vector and based on the voltage tracks value of WAMS. This paper presents a method which is based on the characteristics of the disturbed trajectory clusters.The method acquires the voltage tracks after the disturbances and defines the29clusters characteristics which reflects the the key information of data. Considering huge data and redundant features might interfere the sample, this paper selects the original features and remove redundant using Relief and Recore algorithms to form the initial data set of the prediction systems.We use Libsvm toolbox as a machine of machine learning and study the parameter optimization method and incremental learning algorithm. We improved the structure of the prediction of transient stability system to accommodate the asymmetric error.In this case, we use three classifiers to train the data in each phase and output the "or" logic. This article verifies the feasibility of this method with10-39New England system simulation built in the PSASP software and the results show that this method can effectively improve the reliability of transient stability prediction. We compare the predicted performance in different amount of samples,different fault types and different time scales and analysis the prediction result in unknown topology and unknown operation mode. If there is interference or the network input node information is not complete, then the WAMS system extracts the incomplete data, the traditional forecasting methods will fail.The method mentioned in this article has unique advantages to the prediction of transient stability of incomplete WAMS information.