Adaptive Robust Optimization with Data-Driven Uncertainty for Enhancing Distribution System Resilience

Shuyi Chen, Shixiang Zhu, Ramteen SioshansiarXiv 2025 · DOI · arXiv

Abstract

Extreme weather events pose significant threats to electricity distribution systems, often resulting in widespread outages and substantial economic losses. Traditional resilience strategies frequently decouple proactive infrastructure hardening from reactive emergency response, or rely on rigid parametric assumptions for outage uncertainty. This paper introduces a tri-level optimization framework that addresses these limitations by jointly planning proactive and reactive measures. Central to this approach is the use of Spatio-Temporal Conformal Prediction to generate "distribution-free" uncertainty sets, providing formal coverage guarantees for weather-driven outages. We demonstrate the model's effectiveness using synthetic simulations and a real-world case study of 2018 snowfall events in Massachusetts. Results indicate that the tri-level model consistently outperforms existing benchmarks, offering a robust yet non-conservative solution for utility planning.

Introduction

The increasing frequency and intensity of extreme weather events, such as the 2018 "bomb cyclones," underscore the urgent need for resilient power grids. Current grid resilience strategies often suffer from two primary shortcomings. First, they typically treat proactive measures (e.g., tree trimming, line undergrounding) and reactive responses (e.g., repair crew dispatch) as independent problems. This decoupling ignores the critical interplay where proactive hardening alters the feasible space for reactive recovery. Second, existing models often rely on historical averages or standard parametric distributions to model uncertainty. Such simplifications fail to capture the non-stationary and high-dimensional nature of extreme weather, leading to either overly optimistic or excessively conservative grid configurations.

Methodology

We propose a tri-level optimization model (Min-Max-Min) to bridge the gap between planning and response:

  1. Upper Level (Min): The system planner determines proactive hardening investments under a constrained budget.
  2. Middle Level (Max): An adversarial "nature" selects the worst-case outage scenario from a data-driven uncertainty set.
  3. Lower Level (Min): The planner executes reactive emergency responses to minimize total outage costs given the hardening and realized outages.

The core innovation lies in the Spatio-Temporal Conformal Prediction framework. Unlike traditional machine learning point estimates, conformal prediction produces uncertainty sets that are guaranteed to contain the true outage values with a pre-specified probability. This method is "distribution-free," making it highly adaptable to changing weather patterns and noisy datasets. To solve this complex tri-level problem, we employ duality theory to collapse the lower levels and apply an iterative Benders’s decomposition algorithm.

Experimental Evaluation

The framework was validated using two distinct datasets:

  • Synthetic Simulation: A 10-region system where outages were modeled using SIR (Susceptible-Infected-Recovered) dynamics to simulate the spatial propagation of failures.
  • Massachusetts Case Study: Real-world data from three major 2018 snowfall events across 14 counties. This included NOAA weather data (wind speed, temperature), US Census socioeconomic data, and official outage reports.

Key Results and Discussion

Performance Gains through Joint Optimization

The tri-level model demonstrated superior performance by effectively balancing proactive and reactive strategies. By considering the worst-case scenario during the planning phase, the model ensures that infrastructure hardening is targeted at regions where reactive response would be most constrained.

Performance comparison of the proposed tri-level model against benchmarks across different system sizes (n).

As shown in Figure 1, the tri-level model consistently results in lower costs (indicated by the positive percentage difference of benchmarks) across various system sizes. The "Proactive-Only" and "Co-Optimized" benchmarks fail to account for the adversarial nature of outages as effectively as the proposed framework.

Computational Scalability

A critical requirement for utility-scale planning is computational efficiency. The Benders’s decomposition approach proved to be highly scalable.

Computational efficiency of the Benders’s decomposition algorithm as a function of system size (n).

Figure 6 illustrates that the solution time remains exceptionally low—well under 0.015 seconds—even as the number of regions increases. This efficiency suggests that the model can be readily applied to large-scale distribution networks without significant computational overhead.

Conclusion

This research provides a robust, data-driven framework for enhancing distribution system resilience. By integrating conformal prediction with tri-level optimization, we offer a method that provides formal reliability guarantees without the pitfalls of traditional parametric modeling. Future work will explore the inclusion of strategic human adversaries and broader social equity metrics to ensure that resilience investments benefit all communities equitably.

Summary updated

Abstract

Extreme weather is straining electricity systems, exposing the limitations of reactive responses, and prompting the need for proactive resilience planning. Most existing approaches to enhance electricity system resilience employ simplified uncertainty models and decouple proactive and reactive decisions. This paper proposes a novel tri-level optimization model that integrates proactive actions, adversarial disruptions, and reactive responses. Conformal prediction is used to construct distribution-free system-disruption uncertainty sets with coverage guarantees. The tri-level problem is solved by using duality theory to derive a bi-level reformulation and employing Bender's decomposition. Numerical experiments demonstrate that our approach outperforms conventional robust and two-stage methods.
BibTeX
@misc{chen2025adaptive,
  author = {Shuyi Chen and Shixiang Zhu and Ramteen Sioshansi},
  title = {{Adaptive Robust Optimization with Data-Driven Uncertainty for Enhancing Distribution System Resilience}},
  year = {2025},
  eprint = {2505.11627},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2505.11627},
}
Figure