Study On Static Voltage Stability Margin Prediction Method Based On CNN And XGBoost

With the rapid development of economy and the constraints of social environmental factors,the operating conditions of power system are getting closer to the limit of stability,and new requirements are put forward for the evaluation of voltage stability of power system.The fast calculation of static voltage stability margin is the basic requirement to evaluate voltage stability on-line,the traditional method is difficult to meet the needs of online applications because of the long calculation time and modeling difficulties,and with the wide application of synchronous phase measurement device in the power system,the real-time grid data obtained provides the possibility for the online monitoring of static voltage stability margin.In this paper,the application of extreme gradient boost(e Xtreme Gradient Boosting,XGBoost)and convolutional neural network(CNN)in static voltage stability margin prediction is studied to further improve the accuracy of model prediction,the results are as follows:(1)In the field of data acquisition,a batch acquisition program for samples has been developed.Using continuous current method to track the PV curve,the effect of generator reactive constraint and grid frame structure on static voltage stability margin was studied,and a large number of static stability limit current data of IEEE-39 node system were compiled,and the diversity of samples was guaranteed by random initializing the operating state of the power grid and taking into account different N-1faults.(2)The method of predicting static voltage stability margin based on XGBoost is studied.The predictive performance of the model is improved by adjusting the parameters,the reactive reserve of the generator is more important when analyzing the characteristic importance,the generalization ability of the model can be improved to some extent by selecting the characteristic training model with higher importance,and the prediction performance of the XGBoost algorithm is good by comparing the prediction value and the real value.(3)The static voltage stability margin prediction method based on CNN is studied,the information matrix is optimized,and the network structure and loss function are optimized.Experiments show that by optimizing the information matrix,CNN's input can contain more network topology information and improve the prediction accuracy of the model,adding the BN layer after the convolution layer and improving the loss function will contribute to the further improvement of model performance.The R~2indicator of the improved CNN model increased by 0.0372 and the MAPE indicator decreased by 4.57%.(4)In view of the prediction of static voltage stability margin,the CNN-XGBoost model is proposed,abstract features are extracted by CNN at the bottom of the network,and the extracted features are fitted by XGBoost algorithm.Comparing the models on the same test set,the results show that the prediction results and actual calculation series of each model are basically the same,and the prediction performance of the CNN model,which can extract network topology information,is better than that of the XGBoost algorithm,compared with the highest prediction accuracy of the CNN-XGBoost model.