A Deep Learning-Based Method for Power System Resilience Evaluation

Power system resilience is vital to modern society, as outages caused by extreme weather can severely disrupt communities. Existing statistical and simulation-based methods for resilience quantification are either retrospective or rely on simplified physical models, limiting their applicability. This paper proposes a deep learning-based framework that integrates historical outage and weather data to predict event-level resilience, measured using the resilience trapezoid method. The trained model is then applied to a benchmark weather dataset to estimate regional resilience, with optional socioeconomic and demographic factors incorporated as weighting terms when policymakers wish to emphasize the needs of specific population groups. The effectiveness of the framework is first validated on simulated outage records, showing strong agreement between predicted and simulated resilience values. It is then applied to real historical outage data to assess the resilience of actual power systems. Beyond evaluation, the results can guide targeted investments in distributed energy resources to improve resilience in vulnerable regions.

💡 Research Summary

The paper presents a deep‑learning‑based framework for evaluating power‑system resilience that bridges the gap between traditional statistical and simulation‑based approaches. Recognizing that statistical methods are limited by the scarcity of low‑probability, high‑impact (LPHI) events and that simulation methods require detailed network topology and fragility models, the authors propose to learn the relationship between historical outage records, weather conditions, and system performance directly from data.

The methodology proceeds as follows: (1) Historical outage data (DOE 417, EAGLE‑I) are merged with high‑resolution weather observations (NOAA, ERA5) and normalized system descriptors (e.g., total customers, installed capacity). (2) A benchmark weather dataset containing a representative set of hazardous events is constructed; this same set is used to evaluate every utility under study, ensuring fair cross‑regional comparisons. (3) An encoder‑decoder deep neural network is trained to predict the time‑varying performance curve f(t) for a given event and system. The encoder compresses weather and system features into a latent representation, while the decoder reconstructs the normalized fraction of customers served over time. (4) The predicted curve is integrated using the resilience‑trapezoid formula R = ∫_{T1}^{T2} f(t) dt / (T2‑T1) to obtain an event‑level resilience value Rs,i,k.

Two aggregate metrics are then derived: an unweighted resilience Ru_k, which is simply the average of Rs over all benchmark events, and a weighted resilience Rw_k that incorporates 15 socio‑economic and demographic factors (e.g., households without vehicles, disability prevalence, limited English proficiency). The weighting follows Rw = Ru