Research On Transient Stability Prediction Of Power System Based On Deep Learnin

The safe and stable operation of power systems is closely related to national production and life,which is a solid foundation for the development of modern society.In the context of expanding scale of power grid and increasing intrinsic complex characteristics,fast and accurate transient stability prediction is facing a great challenge.The traditional method of transient stability prediction based on physical modeling has been difficult to meet the prediction requirements of complex and changing power systems.Supported by the rapid development of computer hardware technology and big data technology,data-driven modeling-based artificial intelligence algorithms have made a breakthrough and provided new ideas for power system transient stability prediction.In this paper,we study the method of power system transient stability prediction based on the deep learning theory in artificial intelligence algorithm,and the main research work is summarized as follows:(1)An introduction is made to the mechanism of power system transient stability,and the power system transient stability prediction is considered as a binary classification problem in deep learning.For the construction of the sample data set,the power system time-domain simulation software PSS/E is used for batch simulation,and the basic data set containing transient stability information is generated by setting different fault operation conditions.(2)Based on deep learning for transient stability prediction research,a prediction model of deep residual shrinkage network that can overcome noise interference is constructed.The deep residual systolic network overcomes noise interference by adding a systolic module to the residual unit to make the model more robust.Combined with the characteristics of the power system transient prediction problem,the focal loss function is introduced to solve the problems caused by unbalanced training samples and difficult samples in the training process,and a more comprehensive and comprehensive evaluation index is designed.(3)A transient stability prediction method for power system combined with continuous learning for updating is proposed for the situation where the prediction accuracy decreases to meet the requirements due to major changes in the operation mode and topology of the power system.Experimental results tested on the improved New England 10-machine 39-node system as well as the IEEE 118-node system show that,compared with the commonly used fine-tuning update method,the continuous learning method proposed in this paper can overcome the catastrophic forgetting problem of the model during update under the condition of using only the new scenario data for update.The updated prediction model can meet the prediction requirements in both old and new scenarios,and the scene coverage of the prediction model is improving.