Sieving out Unnecessary Constraints in Scenario Optimization with an\n Application to Power Systems

Miguel Picallo, Florian Dörfler

Abstract

Scenario optimization is a powerful tool for solving chance-constrained problems by using data samples to ensure feasibility. However, standard approaches often lead to overly conservative solutions that sacrifice performance for redundant safety. This paper introduces a novel "sieving" methodology that identifies and removes non-critical constraints after observing the problem's structure. By utilizing a new risk bound based on the number of support constraints, the approach reduces conservatism while maintaining rigorous probabilistic guarantees. We demonstrate its effectiveness in an optimal power flow problem, showing significant improvements in renewable energy integration within distribution grids.

Introduction

In complex engineering systems, such as electrical power grids, optimization must account for inherent uncertainties in demand and generation. Chance-constrained optimization addresses this by ensuring constraints—like voltage limits—are met with a high probability. The scenario approach provides a data-driven framework for these problems, but its "one-size-fits-all" sample size requirements often result in solutions that are "too safe," leading to inefficiencies.

The Sieving Methodology

The authors propose a "wait-and-judge" approach to scenario discarding. Instead of pre-determining the number of scenarios to ignore, the "sieving" algorithm iteratively removes the most limiting constraints based on the observed number of support constraints ($k$). Support constraints are the specific scenarios that define the optimal solution; a low $k$ relative to the number of decision variables indicates a high degree of conservatism.

The core contribution is a new mathematical bound for risk, $\epsilon(k, R)$, which allows for the removal of $R$ constraints post-hoc while still guaranteeing that the probability of violation remains below a target threshold.

Risk bounds as a function of support constraints k for various numbers of discarded scenarios R.

Application to Power Distribution Grids

The method was tested on the IEEE 123-bus test feeder, a standard benchmark for distribution systems. In this context, the goal is to maximize renewable energy injection without violating voltage constraints (typically 0.95 to 1.05 p.u.). Load and generation profiles were derived from real-world datasets to provide a realistic testing environment.

Key Results

The sieving approach (referred to as the "New Scenario" approach) consistently outperformed standard scenario methods. In the power grid simulation, it allowed for a 1% to 3% increase in renewable energy hosting capacity throughout the day.

Comparison of renewable energy injection scenarios over a day, showing the 'New scenario' (sieving) consistently outperforming the 'Standard scenario'.

While a naive "Expectation" approach (which ignores uncertainty) frequently led to voltage violations, the sieving method remained strictly within safety limits. This demonstrates that the number of support constraints can act as an effective "sensor" for grid conservatism, allowing operators to safely trade excess safety margins for improved economic and environmental performance.

Conclusion

By sieving out unnecessary constraints, scenario optimization becomes a much more flexible tool for real-world engineering. This approach provides a path toward more efficient power systems that can accommodate higher levels of renewable energy without compromising reliability. Future work will explore tighter risk bounds and applications to even larger-scale, real-time grid management.

Summary updated

Abstract

Many optimization problems incorporate uncertainty affecting their parameters\nand thus their objective functions and constraints. As an example, in\nchance-constrained optimization the constraints need to be satisfied with a\ncertain probability. To solve these problems, scenario optimization is a well\nestablished methodology that ensures feasibility of the solution by enforcing\nit to satisfy a given number of samples of the constraints. The main\ntheoretical results in scenario optimization provide the methods to determine\nthe necessary number of samples, or to compute the risk based on the number of\nso-called support constraints. In this paper, we propose a methodology to\nremove constraints after observing the number of support constraints and the\nconsequent risk. Additionally, we show the effectiveness of the approach with\nan illustrative example and an application to power distribution grid\nmanagement when solving the optimal power flow problem. In this problem,\nuncertainty in the loads converts the admissible voltage limits into\nchance-constraints.\n
BibTeX
@inproceedings{picallo2019sieving,
  author = {Miguel Picallo and Florian Dörfler},
  title = {{Sieving out Unnecessary Constraints in Scenario Optimization with an\textbackslash\{\}n Application to Power Systems}},
  booktitle = {CDC},
  year = {2019},
  doi = {10.1109/cdc40024.2019.9030244},
}
Figure