Resource Library / Discussion Paper / Enhancing Power System Resilience Under Deep Climate Uncertainty: A Minimax Regret Approach Combined with GNN-cGAN and SWITCH
Enhancing Power System Resilience Under Deep Climate Uncertainty: A Minimax Regret Approach Combined with GNN-cGAN and SWITCH
Discussion Paper
6 July 2026
Abstract
The rising frequency and intensity of extreme weather events introduce deep climate uncertainties that fundamentally challenge the resilience of urban power systems. Conventional deterministic planning approaches are ill-suited to capture such uncertainties, often leading to maladaptive investments and system vulnerabilities. This
study develops a data-driven robust planning approach that integrates high-fidelity scenario generation with a two-stage robust optimization framework. We develop a hybrid generative model combining Graph Neural Networks (GNN) and Conditional Generative Adversarial Networks (cGAN) to capture the city’s spatiotemporal dependencies in meteorological variables and produce physically consistent, high-resolution climate scenarios. These scenarios are embedded into a two-stage robust planning model based on the Minimax Regret criterion, which identifies resilient investment portfolios adaptable to diverse future climate trajectories. Applied to Lvliang City, China, the results demonstrate that the proposed approach improves system resilience by avoiding investment lock-in and reducing the risk of performance degradation under extreme conditions. Compared to conventional planning methods, the framework achieves a more balanced trade-off between economic efficiency and climate adaptability.
Summary
As extreme weather becomes more frequent and severe, cities need power grids that can handle unexpected climate shocks, not just the average conditions planners traditionally design for. This paper, focused on Lvliang City in China, tackles that problem with a two-part approach.
- The researchers built an AI model (combining graph neural networks with generative adversarial networks) that learns from 25 years of real weather data and then generates thousands of realistic, high-detail climate scenarios; including extreme heat, cold, wind, and solar conditions that stress-test the grid in physically believable ways.
- They fed these scenarios into a planning method called Minimax Regret, which chooses an investment plan (how much wind, solar, and storage to build) that minimizes the worst-case “regret”: the gap between what you built and what you would have built if you’d known the future perfectly. This avoids two common planning traps: over-investing defensively against every worst case (wasteful) or betting on one climate outcome and getting burned if it’s wrong (risky).
The results show that traditional, single-scenario planning leaves cities vulnerable. Under mismatched climate conditions, over 13% of electricity demand could go unserved. The new approach, by contrast, builds a balanced mix of wind, solar, and storage from the start, sidestepping that lock-in. It came out ahead on cost in about three-quarters of scenarios tested, with average savings of 14% (and up to 56% in the worst-case comparisons), while also being more reliable. This Pareto-dominant outcome is better on both cost and resilience at once, not a trade-off between them.
Presented At/Published In
Draft
Country
China
Associated Resources
Citation
Wang, Jiarui, et al. “Enhancing Power System Resilience Under Deep Climate Uncertainty: A Minimax Regret Approach Combined with GNN-cGAN and SWITCH.” Draft manuscript, July 6, 2026.