Application of Deep Learning-Driven UAV in Power Equipment Fault Location and Treatment

In the research of fault localization and handling for electrical power equipment, the introduction of deep learning techniques, particularly Long Short-Term Memory (LSTM) networks, has significantly enhanced the accuracy and efficiency of fault detection and localization. Aiming to address the issues of gradient vanishing and gradient explosion inherent in traditional methods for long-term sequence prediction, an improved HT-LSTM model is employed. This model integrates memory units, gating mechanisms, and heuristic threshold processing to elevate prediction precision and robustness. Results indicate that the HT-LSTM model outperforms in various key metrics. In terms of accuracy, HT-LSTM achieves 90.20%, markedly surpassing other algorithms. Both the recall rate and F1 score also exhibit impressive performances, reaching 85.50% and 0.88, respectively. Furthermore, HT-LSTM boasts a detection speed of 40 frames per second, notably superior to alternative approaches. Fault localization reveals that HT-LSTM maintains an error range of 0.1 to 0.2 meters across diverse fault types, significantly outperforming Vanilla LSTM, GRU, and BiLSTM. For specific fault types such as short circuits, overloads, equipment aging, line breaks, and insulation failures, HT-LSTM demonstrates heightened detection accuracy and reliability. In conclusion, HT-LSTM, with its superior predictive performance and robustness, provides a reliable safeguard for the safe operation of electrical power systems, facilitating improved equipment maintenance efficiency, faster fault response, and reduced operational risks and maintenance costs within the power grid.