Research on Intelligent Fault Identification and Power Outage Prediction Algorithm of Power Grid Under Extreme Weather Based on Deep Learning

In the context of global climate change, extreme weather poses a serious challenge to the safe operation of the power grid. Traditional fault identification and power outage prediction methods face problems of insufficient accuracy and weak generalization ability due to the difficulty in handling the spatiotemporal coupling of meteorological power grid data and small sample extreme scenarios. Therefore, this study proposes a deep learning based DL-RAEWM model and constructs an end-to-end intelligent monitoring framework of “data augmentation spatiotemporal feature fusion hierarchical prediction”. The model enhances extreme scenario samples by generating adversarial networks, and utilizes graph convolutional neural networks and Transformers to fuse power grid topology spatial correlations and meteorological temporal dynamic features, achieving collaborative modeling of fault type recognition and power outage duration prediction. The experimental results show that in extreme weather scenarios with a scale of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{1 0 0}$</tex> power grids, the model has a power outage detection rate of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{9 8 \%}$</tex>, a fault prediction rate of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{8 5 \%}$</tex>, and an accuracy rate of 95 %, which significantly improves the performance compared to traditional methods. The model inference delay is controlled within 12.3 ms to meet the real-time warning requirements of the power grid. The research results provide high-precision data support for emergency response of the power grid under extreme weather conditions, and its spatiotemporal fusion modeling and data augmentation technology effectively solve the problem of coupling modeling of multi-source heterogeneous data. Future research will focus on identifying composite faults in new energy power grids, deploying lightweight models, and optimizing adaptive warning strategies to further enhance the resilience and proactive defense capabilities of the power grid in extreme environments.