Power System Frequency Situation Prediction Method Based on Transfer Learning

Regional interconnection and large-scale new energy interconnection bring new risks and challenges to frequency stability and control of power grid. The method of frequency situation prediction of power system based on physical model has contradiction between calculation accuracy and calculation speed. While traditional machine learning method is difficult to adapt to the characteristics of fast operation mode of power system and network topology. Especially when the historical data storage of power grid is insufficient and the sample data is small, the traditional machine learning method is difficult to meet the precision requirements of training. In this paper, a method of power grid frequency situation prediction based on transfer learning is proposed. Its main idea is to select eligible samples from data of the source domain system with similar characteristics and transfer them to the target system, so as to increase the number of samples in the training sample set of the target system and improve accuracy and reliability of the prediction model. In addition, the simulation is carried out on the New England 10-machine 39-node system, and results show that the transfer learning algorithm can make full use of existing historical data and significantly improve the accuracy of the prediction model.