Robust Continuous-Time Generation Scheduling under Power Demand Uncertainty: An Affine Decision Rule Approach

Youngchae Cho, Insoon Yang, Takayuki Ishizaki

Introduction

Traditional power grid operations rely on discrete-time scheduling models, typically balancing supply and demand in hourly or sub-hourly intervals. While these models are computationally manageable, they are increasingly inadequate for modern grids characterized by high penetrations of volatile renewable energy. In a discrete framework, a schedule may appear balanced at the start and end of an hour, yet demand fluctuations within that interval can exceed a generator's physical ramp-rate limits. Such intra-interval imbalances lead to frequency instability and potential blackouts.

This paper addresses the limitations of existing continuous-time models, which often utilize Bernstein polynomials to approximate demand trajectories. While mathematically elegant, these polynomial approaches suffer from a "non-anticipativity" flaw: the coefficients required to determine generator output at a specific time $t$ often depend on demand data from the entire planning horizon. In practice, system operators cannot "see the future," making such models unimplementable in real-time dispatch. This research proposes a robust optimization framework that ensures continuous-time feasibility while strictly adhering to non-anticipative decision-making.

Methodology

The authors introduce a Robust Continuous-Time Generation Scheduling model centered on a novel application of Affine Decision Rules (ADR). The methodology is built upon three primary innovations:

1. Demand Trajectory Uncertainty Sets

Rather than attempting to model every possible demand curve, the authors define a robust uncertainty set $\Omega$. This set is bounded by piecewise constant functions and constrained by specific ramp-rate limits, capturing the worst-case fluctuations that the system must withstand.

2. Non-Anticipative Affine Decision Rules

To ensure the model is physically implementable, the generator output $x$ at time $t$ is defined as a linear function of the current demand $\xi(t)$: $$x(\xi, t) = \alpha \xi(t) + \beta(t)$$ In this formulation, the coefficient $\alpha$ remains invariant over the planning horizon, while the constant term $\beta(t)$ is a continuous piecewise affine function. This structure ensures that the dispatch decision at any moment $t$ depends only on the realized demand at that specific moment, effectively eliminating the "cheating" inherent in polynomial-based approximations.

3. Reformulation and the Cutting-Plane Method

The robust formulation initially presents an infinite-dimensional problem due to the continuous nature of time and uncertainty. The authors reformulate this into a finite-dimensional linear program. To solve it efficiently, they employ a cutting-plane algorithm. This iterative process identifies the most "violated" constraints—points in time where the ramp-rate or capacity limits are most likely to be breached—and adds them to the optimization until a feasible, optimal solution is reached.

Experimental Setup

The proposed model was validated using a modified six-bus test system, a standard benchmark in power systems engineering. The test environment included:

  • Generation: Three fossil-fuel units with varying cost structures and ramp-rate constraints.
  • Infrastructure: Seven transmission lines and three primary demand centers.
  • Data: Historical demand patterns were used to scale and construct the uncertainty sets for a 24-hour planning horizon.

Results and Discussion

Computational Performance

The cutting-plane method demonstrated high efficiency, converging in only 8 iterations for a 24-hour horizon with 64 breakpoints ($M=64$). This suggests that the model is viable for practical utility-scale applications where decision-making speed is critical.

Robustness vs. Scenario-Based Optimization

The ADR model was compared against a traditional scenario-based model optimized for a finite set of predicted futures. While the scenario-based approach yielded a lower average cost, it failed to maintain feasibility in all five test cases when subjected to worst-case demand trajectories. Specifically, it consistently violated generator ramp-rate limits. Conversely, the proposed ADR model remained feasible across all possible trajectories within the uncertainty set, providing a guaranteed upper bound for generation costs ($8.63 \times 10^4$).

Continuous Feasibility

The primary advantage of this approach is the guarantee of "continuous feasibility." By accounting for the infinitesimal physical limits of the generators, the model ensures that the system remains stable at every second of operation, not just at the discrete markers used in traditional scheduling.

Conclusion

This research provides a robust framework for power generation scheduling that bridges the gap between theoretical continuous-time modeling and practical, non-anticipative implementation. By utilizing Affine Decision Rules, the authors offer a method to manage the uncertainty of modern power loads without violating physical hardware constraints.

Future work will focus on integrating Unit Commitment—the binary decisions regarding which plants to turn on or off—which remains a significant combinatorial challenge in continuous time. Additionally, the authors suggest that further research into the systematic construction of uncertainty sets from historical data and the adaptation of electricity markets to continuous-time rules will be essential for the real-world adoption of these methods.

Summary updated

BibTeX
@inproceedings{cho2025robust,
  author = {Youngchae Cho and Insoon Yang and Takayuki Ishizaki},
  title = {{Robust Continuous-Time Generation Scheduling under Power Demand Uncertainty: An Affine Decision Rule Approach}},
  booktitle = {CDC},
  year = {2025},
  doi = {10.1109/cdc57313.2025.11312344},
}
Figure