Fault diagnosis of high-voltage transmission lines based on sequence components and multi-scale spatiotemporal awareness

The transient signals of faults in high-voltage transmission lines usually exhibit strong nonlinearity and inter-phase coupling characteristics. Traditional protection algorithms are prone to false actions or false alarms in conditions with weak characteristics such as high-resistance grounding. Therefore, this paper proposes a fault diagnosis method that integrates physical machine decoupling and multi-scale spatiotemporal perception networks. Firstly, by combining the sliding Fourier transform and the symmetrical component method, independent time-series component features are constructed to eliminate inter-phase electromagnetic coupling. Secondly, a multi-scale one-dimensional CNN network is used to extract transient broadband features, and a BiGRU network with a sliding time-series window is utilized to track the sequence evolution process. The weighted attention mechanism is combined to dynamically focus on the key fault wavefronts. Additionally, the Bayesian algorithm is introduced to achieve global optimization of model hyperparameters. Experimental results based on the 220kV high-voltage three-phase transmission line simulation system show that the model exhibits excellent recognition accuracy in six typical scenarios: no fault, single-phase grounding short circuit, two-phase grounding short circuit, two-phase short circuit, three-phase short circuit, and three-phase grounding short circuit. The model has excellent resistance-to-high-resistance capability and has broad application prospects in smart grid engineering.