A data-and-model-driven approach for stochastic failure probability prediction of overhead transmission lines under typhoon-landslide disaster chains

In addition to the direct impacts of strong wind and heavy rainfall, typhoons can induce secondary geohazards such as landslides, thereby initiating typhoon-landslide disaster chains that pose significant risks to overhead transmission lines located in mountainous and coastal regions. Therefore, this paper proposes a data-and-model-driven framework for stochastic failure probability prediction of overhead transmission lines under typhoon-landslide disaster chains. First, the data-driven component develops a Wasserstein Generative Adversarial Network with Deep Ensemble Learning (WGAN-DEL) model to identify high-risk overhead transmission lines under imbalanced sample conditions. The WGAN generates landslide data to mitigate the imbalanced sample issue due to limited historical data, while the DEL achieves landslide susceptibility assessment to quantify the stochastic spatial distribution uncertainty of typhoon-landslide disaster chains. The data-driven component can thereby identify the high-risk overhead transmission lines. Second, the model-driven component develops a multi-load spatial coupling failure model to quantify the failure probability of high-risk overhead transmission lines identified in the data-driven component. By coupling wind load, landslide load, and soil reaction force load, a multi-load spatial coupling force model for overhead transmission lines is developed. Then, the generalized stress-strength interference theory is applied to quantify the failure probability of overhead transmission lines. Last, a case study on the power grid of Zhanjiang City, Guangdong Province, China, affected by Typhoon Talim (2023), validates the proposed model. Results demonstrate that the model effectively identifies the high-risk overhead transmission lines, and provides theoretical support for disaster prevention and mitigation of overhead transmission lines.