Power system operators face mounting challenges during extreme cold weather when wind generation plummets while heating demand peaks—a scenario increasingly common in grids with high wind penetration. A new research paper tackles this gap by proposing a sophisticated two-stage optimization framework designed to minimize both financial risk and blackout probability during such events.
The approach begins with a machine learning component: a Wasserstein distance-based generative adversarial network (WGAN) produces realistic scenarios of wind power output under cold-wave conditions, capturing the extreme variability that deterministic forecasts miss. This scenario generation feeds into the core optimization strategy, which separates decision-making into two phases. The pre-dispatch phase sets generator schedules and reserves ahead of time, while a rolling dispatch phase continuously reoptimizes unit commitment as real-time conditions emerge.
The crucial innovation lies in using conditional value-at-risk (CVaR), a financial risk metric, to quantify the system's exposure to shortfall costs during rolling dispatch. Rather than assuming worst-case scenarios, CVaR focuses on the tail risk—the average cost if the system enters its worst 5 to 10 percent of outcomes. This more nuanced approach allows operators to hedge against extreme events without over-provisioning expensive backup resources.
Testing on an enhanced IEEE 24-bus test system demonstrated measurable improvements. The two-stage risk-averse method reduced total system costs and unmet demand compared to purely deterministic scheduling, deterministic rolling dispatch, and non-risk-aware approaches. The method balanced operational economy with reliability in ways existing techniques could not achieve.
As cold waves grow more severe and wind resources more critical to decarbonization goals, this framework offers grid operators a practical tool for managing the inherent tension between renewable intermittency and extreme weather demand spikes—a problem that will define grid planning for decades.
