Power System Transient Stability Assessment Based On Deep Learning

With the rapid development of power system.large-scale renewable energy access and UHV ac-dc hybrid has become a trend,the power system becomes more and more complex,and its safety and stability control is also facing a severe test,so it is urgent to explore a method that can carry out rapid and accurate transient stability assessment to provide sufficient response time for emergency preventive measures.In recent years,with the development of artificial intelligence technology,many transient stability assessment methods based on machine learning have been proposed,to a certain extent,these methods overcome traditional methods’ shortcomings that can not effectively take into account accuracy and rapidity.However,since most machine learning models belong to shallow models,their data processing capability is limited in the face of large amounts of running data,and there is still a huge room for improvement in the assessment accuracy and response speed.Aiming at the shortcomings of the existing methods,this paper innovatively introduces the deep learning neural network into the transient stability assessment,which provides a new idea for rapid and accurate assessment.A transient stability assessment method based on one-dimensional convolution neural network(1D-CNN)is proposed.This method can directly target the underlying measurement data and automatically extract the useful information contained in it,so as to achieve the purpose of accurately characterizing the transient stability state of the power system.Specifically,this method can process the bus voltage and branch power flow time series data collected by PMU,by using convolution layer and pool layer stacking,effective transient timing features are automatically extracted from them,and then these features are handed over to the full connection layer for transient stability classification,so as to realize the end-to-end "transient timing feature extraction+stability classification" function,which eliminates the need for artificial feature extraction and reduces the using cost.By virtue of the unique network structure of one-dimensional convolution neural network,the proposed method provides the prior information of" the similarity of measurement data points on the timeline is strong.",and improve the performance of the shallow transient assessment model through the end-to-end of two-stage joint optimization.An adaptive transient stability assessment method based on combinatorial multi-scale convolution is proposed.This method forms an adaptive transient assessment logic by defining the incredible decision parameters,and realizes the continuous stability classification with the increasing data of time series by using the same combinatorial multi-scale convolutional neural network.The proposed method uses the combinatorial multi-scale convolution network to extract multi-scale transient features with different time spans,which can effectively improve the sensitivity of the time series increasing data in the process of adaptive evaluation,so that it has the ability to maintain high evaluation accuracy while improving the response speed.The proposed transient stability assessment method based on deep learning is tested on the New England 10-machine 39-bus test system,and the experimental results show that compared with the existing shallow machine learning transient stability assessment method,the proposed methods have better performance in assessment accuracy and response speed.In addition,the two methods have different emphasis on the accuracy and rapidity of the transient stability assessment,which provides a more flexible choice for practical application.