Feature Extraction and Source Identification for Complex Voltage Sag Based on SAE and Softmax Classifier

With the development of power quality monitoring system, the voltage sag monitoring data is becoming larger and larger, which often exist some problems such as incomplete monitoring information and interference signals. This paper proposes an automatic feature extraction and source identification method of complex voltage sag source based on sparse self-encoder (SAE) and softmax classifier. Firstly, the causes of voltage sag are analyzed and summarized. Then the SAE is used to extract the feature of different voltage sag sources, and the depth feature is automatically gotten with the hidden layer of SAE. Finally, the encoder with the softmax classifier and the fine-tune the stacked network outputs the source classification result. The example results show that the proposed method can accurately identify the complex voltage sag source and has good generalization performance and robustness.