Application Of Data-Driven Machine Learning Methods In Transient Stability Assessment Of Power Systems

With the increased integration of renewable energy sources,the safe and stable operation of power systems has ushered in a whole new set of challenges,especially the operational challenges brought about by the reduced inertia of the system,which will have a significant impact on assessing the transient stability of power systems.With the development of new generation technology,the traditional problem of transient stability assessment has become more complex and its importance has become more and more prominent,which has become the focus of power system technology research.Meanwhile,the rise of "big data" in power systems and the development of wide-area monitoring systems will introduce a new paradigm to address these challenges.Traditional methods of power transient stability assessment,such as direct methods and time-domain simulation methods,are no longer able to meet today’s rapidly changing technological needs,and their shortcomings are becoming increasingly evident.Along with the development of artificial intelligence technology,new methods and means have emerged,providing a new vision and thinking for the in-depth study of transient stability problems.Therefore,how to make full use of this advantage and apply it to transient stability assessment becomes the core of this paper.This paper aims to solve the problem of transient stability assessment of power systems from the perspective of data mining and machine learning,as follows:(1)Based on the dynamic simulation of the IEEE-39 node test case power system,a new open data set of time-domain phasor measurement signals is constructed,and the characteristics of the original data are obtained by using stratified sampling,while several model evaluation metrics for transient stability assessment are proposed.(2)Two special Recurrent neural network structures,namely Gated recurrent unit and Long short-term memory classifiers,were used to build a network model for power system TSA,and then the performance of both Gated recurrent unit and Long short-term memory models for power system TSA was investigated,as well as the effect of the choice of network input features on the model.(3)Considering that the main drawback of Long short-term memory and Gated recurrent unit to obtain the raw time series signals is the inability to transfer the learned information between different power system architectures,a complete machine learning model for power system transient stability assessment is proposed,which is constructed by an undercomplete noise reduction stacked auto encoder and a voting ensemble classifier,and the paper investigates and discusses the classifier in power system test cases The paper investigates and discusses the application results of this classifier in power system test cases.The tested machine learning approach for the power system transient evaluation problem is promising because of its ability to absorb large amounts of data while retaining the ability to generalize and support real-time decision making.The evaluation method performs well,with its fast and accurate evaluation and good fault tolerance.