Risk-Based Distributionally Robust Optimal Power Flow With Dynamic Line Rating

Cheng Wang, Rui Gao, Feng Qiu, Jianhui Wang, Linwei Xin

TL;DR

This paper introduces a novel "risk-based distributionally robust optimal power flow" (DROPF) approach that accounts for the uncertainties introduced by dynamic line ratings (DLR) and wind power generation. By embedding operational risks into the objective function and using a robust optimization framework, the method aims to improve grid reliability and efficiency, outperforming existing methods in simulations.

What Problem Does This Paper Address?

The increasing integration of wind power into electricity grids presents a significant challenge due to its inherent variability. Furthermore, dynamic line rating (DLR) technology, which allows transmission lines to carry more current under favorable weather conditions, introduces additional uncertainty. Traditional power flow models often rely on static line ratings (SLR), which are conservative and can lead to underutilization of transmission capacity. The core problem this paper addresses is how to optimize power flow in a grid with both uncertain wind generation and dynamic line ratings, while explicitly managing the risks associated with load shedding, wind curtailment, and line overloads. Existing methods often struggle to capture the complex correlations between these uncertainties or are overly conservative.

Methods

The authors propose a Risk-Based Distributionally Robust Optimal Power Flow (W&M-DROPF) model. Key methodological contributions include:

  • Risk Embedding: Operational risks (load shedding, wind curtailment, line overload) are incorporated as penalty terms in the objective function.
  • Distributionally Robust Optimization (DRO): To hedge against uncertainties in wind generation and line ratings, a DRO framework is employed. The ambiguity set, which defines the range of possible probability distributions for these uncertainties, is constructed using:
  • Second-order moment constraints: This captures the correlation structure between wind generation outputs and line ratings.
  • Wasserstein distance: This provides robustness against data perturbations and is well-suited for hedging against high-dimensional data.
  • Conic Program Reformulation: The W&M-DROPF model is reformulated into a tractable conic program using strong duality theory.
  • Approximation Techniques: To reduce computational burden, especially for large-scale systems, two approximation strategies are proposed:
  • A general approximation that reduces the number of semi-definite programming (SDP) constraints.
  • A cruder approximation that formulates distributionally robust problems for individual components, leading to a model that grows linearly with the number of generators and lines.
  • Wasserstein Distance Constrained DROPF (W-DROPF): For very large systems where even the approximated W&M-DROPF might be computationally prohibitive, a simplified W-DROPF model is introduced. This model uses only the Wasserstein distance in its ambiguity set and is reformulated as a convex quadratic programming (QP) problem, making it highly scalable.

Datasets

The paper uses three test systems for evaluation:

  • 5-bus test system: A small-scale system used for initial validation and detailed analysis of different approaches.
  • IEEE 118-bus system: A medium-scale system to demonstrate scalability.
  • Polish 2736-bus system: A large-scale system to test the computational feasibility of the proposed methods.

The data for these systems, including topology, generator/load information, and wind farm locations, are referenced from existing literature ([37], [39]). The paper also describes a process for generating synthetic samples for wind power outputs and line ratings, incorporating correlations and ensuring sample validity.

Key Insights and Ideas

  • DLR's Double-Edged Sword: While DLR significantly increases transmission capacity and reduces dispatch costs, it also introduces substantial uncertainty that needs careful management.
  • Correlation Matters: The correlation between wind generation and line ratings is a crucial factor. Ignoring it can lead to suboptimal or overly conservative solutions. The proposed ambiguity set effectively captures this.
  • Wasserstein vs. Moment Constraints: The Wasserstein distance provides robustness against data perturbations and is generally more effective for hedging against rare events. Second-order moment constraints are good for capturing correlations but can be overly conservative if the underlying distribution is not well-represented by moment information alone.
  • Hybrid Approach is Best: The W&M-DROPF model, combining both Wasserstein distance and moment constraints, consistently outperforms models using only one type of constraint, demonstrating the benefit of a hybrid approach.
  • Approximations are Crucial for Scale: For practical, large-scale power systems, approximations are not just beneficial but necessary to make the problem computationally tractable, even if they introduce some degree of conservativeness.

Key Results

  • Improved Performance: The proposed W&M-DROPF model consistently achieved better out-of-sample performances (lower operational costs under uncertainty) compared to other methods like M-DROPF (moment-based), W-DROPF (Wasserstein-based), and SAA (Sample Average Approximation).
  • DLR Benefits: All tested models showed reduced dispatch costs when using DLR compared to SLR, highlighting the economic advantage of DLR.
  • Scalability: The W-DROPF model, formulated as a QP, demonstrated excellent scalability and could be solved efficiently for large-scale systems (Polish 2736-bus) where the more complex W&M-DROPF and M-DROPF models were computationally intractable.
  • Approximation Effectiveness: The approximated versions of W&M-DROPF and M-DROPF provided significant reductions in computational time, with the linear approximation (16) being particularly efficient. While approximations could slightly worsen performance in some cases, they were crucial for handling larger sample sets.
  • Penalty Coefficient Impact: The study showed that penalty coefficients for risks significantly influence the dispatch costs and reserve allocation strategies, with higher penalties leading to higher costs but better risk mitigation.

Limitations and Future Work

  • Computational Burden of W&M-DROPF: Despite reformulations and approximations, the W&M-DROPF model remains computationally intensive for very large systems, requiring specialized solvers and potentially significant computational resources.
  • Sample Size Dependency: The performance of DRO models can be sensitive to the number of available samples. While the paper explores different sample sizes, the optimal tuning of ambiguity set parameters might still be challenging with very limited data.
  • DC Power Flow Assumption: The study uses a DC power flow model for simplicity. Extending the approach to AC power flow would introduce non-linearities and increase complexity.
  • Future Work: The authors suggest extending the ambiguity set construction to incorporate more sophisticated statistical properties and exploring the application of their framework to other power system decision-making problems like economic dispatch and reserve procurement. The potential for extending the model to multi-period problems is also mentioned.

Summary updated

Abstract

In this paper, we propose a risk-based data-driven distributionally robust approach to investigating the optimal power flow with dynamic line rating. The risk terms, including penalties for load shedding, wind generation curtailment and line overload, are embedded into the objective function. To robustify the solution, we consider a distributional uncertainty set based on the second-order moment, that captures the correlation between wind generation outputs and line ratings, and also the Wasserstein distance, that hedges against data perturbations. We show that the proposed model can be reformulated as a convex conic program. Approximations of the proposed model are suggested, which leads to a significant reduction of the number of the constraints. For practical large-scale test systems, a distributionally robust optimal power flow model with Wasserstein-distance-based distributional uncertainty set and its convex reformulation are also provided. Simulation results on the 5-bus, the IEEE 118-bus and the Polish 2736-bus test systems validate the effectiveness of the proposed models.
BibTeX
@article{wang2018risk,
  author = {Cheng Wang and Rui Gao and Feng Qiu and Jianhui Wang and Linwei Xin},
  title = {{Risk-Based Distributionally Robust Optimal Power Flow With Dynamic Line Rating}},
  journal = {IEEE Transactions on Power Systems},
  year = {2018},
  doi = {10.1109/tpwrs.2018.2844356},
}

Loading related content...

Figure