Research On Evaluation Method Of Power System Transient Stability On Convolution-recurrent Neural Network

The wide use of electric energy makes the interconnection scale of modern power system expand,and the influence range of power outage caused by transient instability is also increasing.The access of a large number of renewable energy and the use of AC / DC hybrid technology increase the probability of transient instability of the system.Therefore,how to quickly and accurately evaluate the transient stability of power system has become a hot topic of relevant scholars.With the rapid development of communication technology,the era of power data has come.Wide application of wide area measurement system provides data support for transient stability assessment method based on system real-time response information,and further extends the transient stability evaluation of power system based on machine learning.However,the evaluation model based on shallow machine learning method has insufficient ability to analyze the characteristics,and there is still room for improvement in the evaluation performance.Therefore,this paper introduces convolutional-recurrent hybrid network into transient stability assessment,which provides a new idea for fast and accurate assessment of the transient stability of the system.The main research contents are as follows :In this paper,through the selection of input characteristics and the construction of the model,a model with good evaluation performance for data containing noise and data imbalance is proposed.Based on the idea of convolutional-recurrent hybrid network,a CNN-LSTM transient stability evaluation model is built.In order to enhance the generalization ability of the model,batch normalization algorithm and attention mechanism are added to the model.At the same time,the introduction of differential evolution algorithm strengthens the model parameter adjustment ability and further improves the model performance.The time series trajectory of generator power angle and rotor angular velocity after fault removal is obtained by simulation setting as input feature,and the data set is divided into training set and test set to complete the model construction.The input feature of the model is the measured data of the bottom generator.Compared with the evaluation model based on shallow machine learning,it does not need to calculate the input feature artificially,but the model independently mines the implicit relationship between transient stability and input.The simulation results of the example show that the performance of the model using the generator power angle and rotor angular velocity as the input feature is better than that using the generator power angle or rotor angular velocity as the input feature;batch normalization and attention mechanism increase the adaptability of the model;this evaluation model still has good robustness in the face of noise and imbalance of input data.