As solar power floods the grid, the "duck curve"—the shape of net electricity demand over a day—has become the defining challenge for power systems engineering. But there is a catch: traditional duck curves are deterministic, showing only a single "typical" or "extreme" day. In reality, weather is fickle, and load is variable.
A team of researchers from Tsinghua University, publishing in Applied Energy, argues that we shouldn't be looking at a single line, but a cloud of possibilities. They propose the Probabilistic Duck Curve (PDC), a framework to help grid planners navigate the inherent uncertainty of high-renewables systems.
TL;DR
Traditional duck curves ignore the massive uncertainty of solar generation, leading to potential over-investment or grid instability. This paper introduces the Probabilistic Duck Curve (PDC), which uses copula functions and discrete convolution to model the full range of net-load scenarios, proving that retrofitting coal plants for flexibility is currently more cost-effective than energy storage for accommodating solar in China.
What Problem Does This Paper Address?
The "duck curve" represents net load (total demand minus renewable generation). It features a deep "belly" at midday when solar is peaking and a steep "neck" (ramp) in the evening as the sun sets and demand rises.
The problem is that standard models are deterministic. They don't tell you how likely a specific dip or ramp is, nor do they account for the correlation between different solar farms. If a planner only looks at an average day, they might be blindsided by a day with extreme cloud cover or unexpected demand spikes. Conversely, planning for the absolute worst-case scenario every day leads to massive, unnecessary costs.
Methods
The authors developed a four-stage modeling framework to move from single lines to probability distributions:
- Marginal Distributions: Instead of assuming solar follows a standard "Beta" or "Normal" distribution, they used Kernel Density Estimation (KDE). This non-parametric approach better fits real-world data, which is often skewed or has "fat tails."
- Dependency Modeling: Solar farms near each other don't operate in isolation; if it's cloudy over one, it's likely cloudy over the neighbor. The authors used Copula functions to model these complex spatiotemporal dependencies between PV sites and between PV and load.
- Dependent Discrete Convolution (DDC): To combine these variables into a single net-load distribution without massive computational overhead, they employed DDC. This allows for efficient calculation of the sum (or difference) of dependent random variables.
- New Indices: They introduced metrics like the Probabilistic Area (S), which quantifies the expected energy curtailment, and the Probabilistic Ramp Curve (PRC) to visualize the uncertainty of evening ramps.
Datasets
The model was validated using an empirical study of the Qinghai provincial power system in China. Qinghai is a "high PV penetration" pioneer; by 2020, its PV capacity reached 10 GW, accounting for roughly 77% of its peak load. The researchers used historical PV generation and load data from the region, specifically focusing on clusters like the Golmud PV farms.
Key Insights and Ideas
- Uncertainty is Time-Dependent: The "belly" of the duck (midday) exhibits much higher uncertainty than the "head" (night). This means flexibility requirements aren't just about the magnitude of the dip, but the variance of it.
- The "Fatter Tail" Effect: By using Copulas to model dependencies, the researchers found that traditional models underestimate extreme events. When PV farms are highly correlated, the risk of a massive, simultaneous drop in generation is higher than independent models suggest.
- The Marginal Benefit of Flexibility: The paper introduces a way to calculate the "marginal probabilistic area." This allows planners to see exactly how much solar curtailment is avoided for every 1 MW reduction in a coal plant's minimum stable output.
Key Results
- The Ramping Challenge: In Qinghai, the evening ramp requirement can reach 4,000 MW/h. The PDC shows that this ramp isn't just steep; it's highly uncertain, with a wide range of possible requirements depending on the day.
- Coal Retrofitting vs. Storage: The study found that reducing the minimum output of coal-fired units (retrofitting for flexibility) is remarkably effective. On average, reducing a coal unit's minimum output by 1 MW increases solar accommodation by over 4 MWh per day.
- Cost-Effectiveness: At current prices in China, retrofitting coal plants is significantly more economical than building new energy storage. The marginal cost of solar accommodation via coal retrofit was calculated at $0.32/MWh**, compared to **$3.97/MWh for 5-hour energy storage.
Limitations and Future Work
The authors acknowledge a few areas for expansion:
- Wind Integration: The current model focuses heavily on PV. Wind power has different temporal patterns (often peaking at night) and would add another layer of complexity to the PDC.
- Copula Selection: While they used Gaussian Copulas, other types (like t-Copulas) might better capture "tail dependence"—the tendency of extreme events to happen together.
- Geographic Scale: The study is provincial. Future work could examine how interconnecting different provinces (spatial smoothing) reduces the overall "probabilistic area" of the duck curve.
Reference: Hou, Q., Zhang, N., Du, E., Miao, M., Peng, F., & Kang, C. (2019). Probabilistic duck curve in high PV penetration power system: Concept, modeling, and empirical analysis in China. Applied Energy, 242, 205-215.
Summary updated