Distribution network fault comprehensive identification method based on voltage–ampere curves and deep ensemble learning

•The three-phase and zero-sequence voltage-ampere curves are used to identify fault line, type, cause and section.•Multimodal residual network model based on RGB normalization is designed to extract the voltage-ampere curves.•Based on fault section identification, the fault phase voltage-ampere curves are used to estimate precise fault distance.•Multimodal residual network model based on vertical segmentation technique is designed to extract the fault phase V-I curve.•The deep ensemble learning model enable comprehensive fault diagnosis in different situations. To identify and locate faults of small-current grounded distribution networks under high-impedance fault with weak characteristics, a fault comprehensive identification method for distribution networks based on voltage-ampere curves and deep ensemble learning is proposed. First, the correlations of the voltage-ampere curves with the fault causes, fault types, and fault distances are analyzed to illustrate the feasibility of using three-phase and zero-sequence voltage-ampere curves as input features. In addition, a multimodal residual network model is developed to extract the fault features using RGB normalization and an attention mechanism. Moreover, a vertical segmentation technique is employed to enhance the feature extraction for fault location by using fault-phase voltage-ampere curves at the identified section to improve the overall fault comprehensive identification performance. Finally, the advantages of the proposed fault identification model and fault location model are validated through comparison experiments. Moreover, the proposed method has significant advantages over the impedance method and artificial neural network method for fault section identification and fault distance estimation. The proposed method has good adaptability and generalization ability to different systems and real-world data. The proposed method can provide decision guidance for automatic line reclosing, fault recovery and operation and maintenance repair. © 2017 Elsevier Inc. All rights reserved.