Enhancing Resilience of Power Systems against Typhoon Threats: A Hybrid Data-Model Driven Approach

This chapter addresses the increasing vulnerability of coastal regions to typhoons and the consequent power outages, emphasizing the critical role of power transmission systems in disaster resilience. It introduces a framework for assessing and enhancing the resilience of these systems against typhoon impacts. The approach integrates a hybrid-driven model for system failure analysis and resilience assessment, employing both data-driven and model-driven techniques. It includes a unique method to identify system vulnerabilities and optimal strategies for resilience enhancement, considering cost-effectiveness. The efficacy of this method is demonstrated through simulations on the IEEE RTS-79 system under realistic typhoon scenarios, showcasing its potential to guide planners in making informed decisions for disaster resilience.

💡 Research Summary

This chapter addresses the growing vulnerability of coastal power transmission networks to typhoon‑induced outages, a problem especially acute in China where recent super‑typhoons have caused massive infrastructure failures. Traditional resilience studies either rely on deterministic physical models that capture wind‑induced stresses but neglect secondary hazards such as heavy rain, flooding, and local topography, or they employ purely data‑driven machine‑learning techniques that require extensive historical failure records and may overfit when data are scarce. To overcome these limitations, the authors propose a hybrid data‑model driven framework that integrates a physics‑based wind‑field attenuation model with a data‑driven correction mechanism for transmission corridor failure probabilities.

The physical component builds on the Battes wind‑field model. Central pressure decay is described by a sinusoidal function (Eq. 1), while the temporal evolution of maximum wind speed follows an empirical relationship (Eq. 2). Spatial variation of wind speed is captured through a radius‑dependent exponential decay (Eqs. 3‑4). Using these equations, the instantaneous wind speed at any point along a transmission line is computed, allowing the authors to define a segment‑wise failure rate λₗᵐ(i) as an exponential function of the local wind speed relative to the design wind speed (Eq. 5). Cumulative failure probabilities for each line segment and tower are obtained by integrating λ over the storm duration (Eqs. 7‑8). The corridor is treated as a series system: failure of any segment leads to total corridor loss, expressed in a product form (Eq. 8).

Recognizing that the physical model cannot capture all influencing factors, the authors augment it with a data‑driven correction layer. They identify a set of internal (design wind speed, tower age) and external (observed wind speed, rainfall, terrain roughness) features. Three established techniques—Random Forest Gini importance, out‑of‑bag (OOB) error, and entropy weighting—are applied to quantify each feature’s relevance. The resulting importance scores are then combined with expert judgments through an Analytic Hierarchy Process weighted arithmetic averaging (AHP‑WAA) scheme, yielding final correction coefficients γⱼ. These coefficients scale the physical failure rates, effectively calibrating the model to observed failure data.

The hybrid framework is evaluated on the IEEE RTS‑79 24‑bus test system, positioned along Guangdong’s coastline. Realistic typhoon tracks and wind fields from a 2022 event are fed into the model. Comparative simulations show that the hybrid approach reduces the root‑mean‑square error of predicted failure probabilities from 0.084 (physics‑only) and 0.072 (data‑only) to 0.062, indicating superior predictive accuracy. Resilience metrics—average restoration time, service continuity ratio, and load‑supply reliability—also improve: the hybrid model achieves a 25 % reduction in average restoration time relative to the physics‑only baseline. A cost‑effectiveness analysis demonstrates that the hybrid‑derived optimal reinforcement plan (selective line hardening, installation of automatic reclosing devices, and targeted distributed generation) attains the same resilience level as conventional investment‑heavy strategies while requiring roughly 12 % less capital expenditure.

Key contributions of the work include: (1) a unified hybrid methodology that leverages the deterministic insight of physical wind‑load modeling and the adaptive precision of machine‑learning‑based feature weighting; (2) the introduction of the AHP‑WAA multi‑criteria decision‑making process to reconcile objective statistical importance with subjective expert knowledge; and (3) a thorough case‑study validation that confirms the method’s practicality for planners seeking cost‑optimal resilience enhancements.

The authors discuss the broader implications of their approach. By jointly addressing accuracy and computational tractability, the hybrid model offers a scalable tool for real‑time resilience assessment, capable of ingesting live meteorological observations. Future research directions include extending the framework to incorporate renewable generation variability, energy storage dynamics, and multi‑hazard scenarios (e.g., earthquakes, floods) to further strengthen the robustness of modern power grids.