Research On Prediction Method Of Power System Transient Stability Based On Deep Learning

With the rapid development of modern power systems,new energy resources are integrated into the power grid on large scale,the trend of UHV AC/DC hybrid connection is becoming more and more obvious,and the interconnection degree of power grids is becoming closer and closer.As a result,the scale and complexity of the power system continue to expand,and it is getting closer to the limit of safe and stable operation.In addition,modern power systems have high-dimensional nonlinearity,fast fault occurrence,and short response time,which undoubtedly increases the difficulty of transient stability prediction.In order to leave sufficient time margin for emergency control after faults,a more rapid and accurate transient stability prediction method is urgently needed.In recent years,the field of artificial intelligence represented by data mining and machine learning has developed rapidly.With the rapid development of the Wide-Area Measurement System(WAMS)based on the Phasor Measurement Unit(PMU)Popularization provides massive measurement data,making data-driven artificial intelligence prediction methods that are more capable of both speed and accuracy than traditional power system transient stability methods have become a current research hotspot.Although this kind of method has many advantages,it still has improvements in many aspects such as model selection and feature construction.Aiming at the shortcomings of existing methods and in order to further improve the speed and accuracy of prediction,this thesis studies the transient stability prediction method of power system based on deep learning theory.The main work of this thesis is as follows:In order to eliminate the uncertainty of artificial input features and improve the accuracy of transient stability prediction,an end-to-end transient stability prediction method based on two-dimensional convolutional neural network was proposed.Firstly,the original input features are selected according to the generator node data which can be measured by PMU.Then,considering the high dimensional timing sequence of the input characteristic data,two-dimensional convolutional neural network was used to construct the transient stability prediction model.Finally,according to the idea of recursive feature elimination,the optimal input feature combination is selected,and the optimal prediction model is obtained by training to predict the transient stability of the power system.The proposed method utilizes the excellent feature self-extraction capabilities of the two-dimensional convolutional neural network to fully mine the timing information of a single feature and the associated information of different features,effectively solving the difficulty of input feature construction in the traditional data-driven power system transient stability prediction,realizes the end-to-end transient stability prediction of power system,and the prediction performance is effectively improved through a reasonable and systematic feature selection method.In order to solve the incompatibility problem of model prediction accuracy and speed under multiple sampling points,a transient stability prediction method based on long short-term memory network combined two-dimensional convolutional neural network was proposed.In this method,the long and short term memory network is used for feature time series prediction,and on this basis,the two-dimensional convolutional neural network which uses PMU data and feature time series prediction data as its input and outputs power system transient stability prediction results is superimposed.The proposed method takes full account of the positive effect of increasing sampling time on the prediction accuracy,gives full play to the time series prediction ability of the long and short-term memory neural network and the feature self-extraction ability of the two-dimensional convolutional neural network,and realizes the transient stability prediction of power system with high speed and accuracy under less data requirements.The New England 10-machine 39-bus system is used as the test system,and appropriate evaluation indicators are selected to evaluate the proposed transient stability prediction method.The experimental results show that the proposed method has better prediction performance compared with the previous power system transient stability prediction methods based on artificial intelligence.