Assessment Of Power System Transient Stability Based On Machine Learning

The safe and stable power system is the important foundation and support for the stable development of modern society.With the continuous expansion of the power system scale,the access of various new energy scomplexity of power grid operation,which is facing a severe test.Power systems are ources,new types of loads,and the wide application of power electronic devices have increased the uncertainty and often subject to disturbances.When large disturbances occur,such as short circuits,wire breaks,or non-fault-trip,it is necessary to study the transient stability of the power system.With the rapid development of computer technology and artificial intelligence,big data and machine learning method is more and more applicated in power system analysis.The machine learning views that there is a mapping relationship between the transient stability of the power system and some of the feature quantities which describe the system operating state.By establishing the model and training it offline with the historical data,analyzing and extracting the unknown function relations,and then collecting the real-time on-line operation data to continuously updating the structure and parameters of the model,the discrimination of the transient stability of power system can be achieved.Based on the machine learning perspective and the transient stability assessment process,the research mainly includes the following aspects:(1)The power system simulation software PSAT is used to carry out the simulation operation,the fault data under different operating conditions is collected,and a set of fault feature quantities is constructed,which strongly related to transient stability and do not vary with the size of the system.The initial sample set for transient stability assessment of power system is obtained by processing system fault feature quantities.(2)The support vector machine(SVM)algorithm is adopted to assess the power system transient stability,because the SVM has been proved to have good performance in the related research.On this basis,a support vector machine model based on Mahalanobis distance is constructed.The Mahalanobis distance is applied to the kernel function to solve this model.Then the sample set with fault feature quantity is trained and the test set is carried out.The assessment verifies the effectiveness of the proposed method.(3)Machine learning is a shallow learning model,its algorithm improvement effect and the ability to extract data features is limited.Therefore,deep learning is introduced to make a better representation of data feature laws.The combination of stacking automatic encoder and support vector machine algorithm is constructed to train the sample set and test the accuracy of the model.Considering that the stacking auto encoder is the most basic deep learning model,the deeper and more complex convolutional neural network is applicated in the model for the assessment of power system transient stability,furthermore,the support vector machine algorithm is also used in the discriminating mechanism of the output layer.The simulation results verify the two proposed models are effective and the performance of transient stability assessment is improved.