Large-scale Grid Optimization: The Workhorse of Future Grid Computations

TL;DR

The paper surveys the rise of large-scale grid optimization, arguing that expanding spatial features (notably DER penetration in distribution) and growing temporal uncertainty (renewables and storage) drive demand for integrated transmission and distribution analyses. It systematically classifies the field with a taxonomy linking application domains to solution techniques, then analyzes advances in physics-based methods for transmission and distribution as well as emerging data-driven approaches that aim to tackle intractable problems. Key findings indicate that physics-based methods still lead the field, while data-driven, especially physics-informed ML, offer promising avenues for scalability and new capabilities such as constraint screening and fast warm-starts. The authors also identify gaps—such as robustness to model error, market-scale coordination, and the need for realistic large-scale test networks—and call for closer industry-academic collaboration to develop benchmarks and open data for validation and benchmarking, enabling practical deployment.

Abstract

Purpose: The computation methods for modeling, controlling and optimizing the transforming grid are evolving rapidly. We review and systemize knowledge for a special class of computation methods that solve large-scale power grid optimization problems. Summary: Large-scale grid optimizations are pertinent for, amongst other things, hedging against risk due to resource stochasticity, evaluating aggregated DERs' impact on grid operation and design, and improving the overall efficiency of grid operation in terms of cost, reliability, and carbon footprint. We attribute the continual growth in scale and complexity of grid optimizations to a large influx of new spatial and temporal features in both transmission (T) and distribution (D) networks. Therefore, to systemize knowledge in the field, we discuss the recent advancements in T and D systems from the viewpoint of mechanistic physics-based and emerging data-driven methods. Findings: We find that while mechanistic physics-based methods are leading the science in solving large-scale grid optimizations, data-driven techniques, especially physics-constrained ones, are emerging as an alternative to solve otherwise intractable problems. We also find observable gaps in the field and ascertain these gaps from the paper's literature review and by collecting and synthesizing feedback from industry experts.

Figures (1)

• Figure 1: Taxonomy for large-scale grid optimizations. The left of the figure describes various facets of grid optimization problem formulation. The right of the figure describes the relevant solution techniques.