Abstract
The high penetration of photovoltaic (PV) generation is reshaping electricity net-load curves, creating significant challenges for power system operation and planning. While the "duck curve" concept is widely used to describe the timing imbalance between peak demand and PV generation, traditional models are deterministic and fail to capture the inherent uncertainty of PV generation and loads. This paper proposes the novel concepts of the Probabilistic Duck Curve (PDC) and Probabilistic Ramp Curve (PRC). Using kernel density estimation, copula functions, and dependent discrete convolution, the authors model the full range of net-load scenarios. An empirical study of the Qinghai provincial power system in China validates the method, demonstrating that retrofitting coal-fired units for flexibility is a highly cost-effective strategy for accommodating high PV penetration.
Introduction
As global power systems transition toward renewable energy, the "duck curve"—characterized by a midday dip in net load followed by a steep evening ramp—has become a central concern for grid stability. However, standard industry practice relies on deterministic duck curves that represent only a single "typical" or "extreme" day.
These deterministic models ignore the massive uncertainty of solar generation caused by weather variability. Relying on a single line can lead to either over-investment in unnecessary capacity or grid instability when actual conditions deviate from the "typical" scenario.
Methodology: From Deterministic to Probabilistic
To address these weaknesses, researchers from Tsinghua University developed a framework to move from single lines to probability distributions. The Probabilistic Duck Curve (PDC) visualizes the net load as a "cloud" of possibilities rather than a fixed trajectory.
The modeling approach involves four key stages:
- Kernel Density Estimation (KDE): Used for marginal distributions to better fit real-world data that may be skewed or have "fat tails," unlike standard Beta or Normal distributions.
- Dependency Modeling: Solar farms near each other don't operate in isolation. The authors used Copula functions to model these complex spatiotemporal dependencies between PV sites and between PV and load.
- Dependent Discrete Convolution (DDC): An efficient computational method to combine these variables into a single net-load distribution without massive computational overhead.
- New Indices: Metrics such as the Probabilistic Area (S) and Probabilistic Ramp Curve (PRC) quantify expected energy curtailment and the uncertainty of evening ramps.
Case Study: The Qinghai Power System
The model was tested on the Qinghai provincial power system, a region where PV capacity reached 10 GW by 2020, accounting for approximately 77% of peak load. The analysis revealed that uncertainty is highly time-dependent; the "belly" of the duck (midday) exhibits much higher variance than the "head" (night). Furthermore, by accounting for dependencies between PV sites, the model showed that traditional methods often underestimate the risk of extreme events, such as simultaneous drops in generation across multiple solar farms.
Economic Implications for Grid Flexibility
A critical application of the PDC is in flexible resource planning. The study compared the cost-effectiveness of retrofitting existing coal-fired units versus installing energy storage.
- Coal Retrofitting: Reducing the minimum stable output of coal units proved remarkably effective. On average, a 1 MW reduction in minimum output increased daily PV accommodation by over 4 MWh.
- Cost Comparison: In the Chinese context, the marginal cost of solar accommodation via coal retrofit was calculated at $0.32/MWh, significantly lower than the $3.97/MWh estimated for 5-hour energy storage.
Conclusion
The Probabilistic Duck Curve provides a more robust foundation for power system planning under high PV penetration. By quantifying uncertainty, grid operators can make more informed decisions regarding flexibility requirements. In the near term, for regions like Qinghai, enhancing the flexibility of existing thermal assets remains a more economical solution than large-scale storage for managing the "duck's" belly and neck. Future work will expand this framework to include wind integration and more complex dependency models.
Summary updated