A data-physics hybrid driven fault identification model integrating multivariate fault information of wind farm

• A multivariate information fusion strategy of wind farm is proposed that can more effectively capture the diverse fault characteristics of the power system. • A data-physics hybrid drive fault identification model to realize the fault identification is proposed, which fully combines the advantages of the data-driven model based on artificial intelligence algorithms and the physical constraints of the power system. • The Phasor-Aware Graph Sampling and Aggregation Network (PAG-SAGE) as the main body of fault identification model is constructed, which contains the real-complex domain dual-domain convolution that ensures the original phasor information representation during the training process and improves the generalization ability of the model, and further achieves fault identification accuracy. • Kirchhoff’s current law (KCL) as the node characteristic physical constraint equation is used and added to the PAG-SAGE model in a regularized manner, in order to realize the physical drive of the model to further correct the parameter optimization trend in the model training process and allow the model to efficiently learn the significant physical characteristics between bus faults and non-bus faults during training. • The proposed model can effectively and precisely identify fault line in special fault cases with weak fault characteristics and in variable topology scenarios, thus avoiding the problem of incorrect activation of traditional protection due to the complex fault characteristics after renewable energy integration, which makes it difficult to properly isolate the faulty line. Due to the fault characteristics of renewable energy source such as limited short-circuit current amplitude and controlled phase angle, the traditional relay protection based on the power frequency short-circuit characteristics of synchronous generator power supplies faces issues such as complex adaptive setting coordination and insufficient interactive coordination of judgment principles. In view of the current situation that traditional rely protection is not adaptable enough after the renewable energy source access, starting from the wind farm information fusion strategy, a data-physics hybrid driven fault identification model with Phasor-Aware Graph Sampling and Aggregation Network as the main body and embedded in the physical prior constraints of the power system is constructed, which takes the real-complex dual-domain feature deep mining of multi-point electrical phasor information of the wind farm and the overall idea of fusion decision of the field global feature, and constructs the graph sampling mechanism and regularization strategy that embeds wind farm topology and physical formulas, in order to further improve the model’s generalization ability and understanding of the physical distribution law of the power system. Finally, the adaptability and effectiveness of the proposed identification model is verified by experimental results, which show that the identification model can improve the accuracy of fault identification under special fault conditions with weak fault characteristics and variable topological conditions. © 2017 Elsevier Inc. All rights reserved.