With the development of industry and science and technology,the user's concern about voltage dips is increasing,and it is more and more important to identify the causes of voltage dips.Aiming at the single and composite sources of voltage dips,this paper proposes a recognition method that combines the S-transform feature extraction and deep learning automatic feature extraction.Firstly,the numerical modeling framework is used to generate mathematical models of single and composite transient sources,and then the transient datasets of nine fault categories are obtained and used as the original data,and secondly,the original data are processed,i.e.,two adjustment factors are introduced to obtain the improved S-transform based on the standard S-transform to obtain the S-transform data,and 16 metrics are introduced to extract features from the S-transform data and are used as the metrics'features.The original data and the S-transformed data are used as model inputs,and the spatial features are extracted from the transient data by using convolutional neural network(CNN),while the data are divided into several one-dimensional vectors and input into bi-directional long short-term memory networks(Bi-LSTM),which is the most suitable method to extract the spatial features of the S-transformed data.Memory networks(Bi-LSTM)to extract temporal features,and finally establish a multi-feature fusion S-CNN-LSTM recognition model with indicator features,spatial features and temporal features.The simulation results show that the recognition accuracies without feature fusion and after multi-feature fusion are 98.36%and 99.08%,respectively,indicating that multi-feature fusion can improve the recognition accuracy.
voltage transient sourceimproved S-transformCNN-LSTM modelfeature fusion
万方数据
