Distribution network operators face persistent challenges in accurately detecting and classifying electrical faults amid complex signal data from multiple monitoring points. Traditional fault identification methods often struggle with distinguishing genuine system failures from harmless operational transients, leading to false alarms and delayed response times. A new research approach addresses these limitations through intelligent signal processing and machine learning optimization.
The proposed method leverages spatiotemporal analysis of travelling wave signals—electromagnetic disturbances that propagate through power lines when faults occur. By extracting statistical features from the time-frequency domain at numerous measurement locations simultaneously, engineers can build a comprehensive picture of how fault signals differ from normal grid fluctuations. Key metrics include modal energy distribution, central frequency characteristics, spectral entropy, frequency change rates, and signal sparsity across the network.
To enhance classification accuracy, the method employs the Whale Optimization Algorithm to fine-tune hyperparameters of XGBoost, a powerful gradient boosting machine learning framework. This hybrid approach, termed WO-XGBoost, dynamically adjusts model parameters during training, improving both detection accuracy and computational speed compared to conventional neural network and traditional statistical methods.
Experimental validation demonstrates superior performance against established fault identification techniques in terms of both accuracy and training efficiency. The methodology shows promise for real-time implementation in modern distribution networks, where rapid fault detection is essential for maintaining service continuity and preventing equipment damage.
For utilities managing increasingly complex grids with distributed generation and microgrids, this advancement offers practical benefits: faster fault isolation, reduced outage duration, and lower operational costs. As distribution networks become more heterogeneous and harder to monitor manually, AI-assisted fault identification represents a critical step toward autonomous, self-healing grid infrastructure.
