Data-Driven Two-Stage Distributionally Robust Dispatch of Multi-Energy Microgrid

Xunhang Sun, Xiaoyu Cao, Bo Zeng, Miaomiao Li, Xiaohong Guan, Tamer BasararXiv 2025 · DOI · arXiv

Introduction

Managing Multi-Energy Microgrids (MEMGs) involves balancing volatile renewable energy (RE) sources with unpredictable consumer demands across electricity, heat, and hydrogen sectors. Traditional optimization methods often struggle with this complexity: Stochastic Programming (SP) requires precise probability distributions that are rarely available, while Robust Optimization (RO) can be overly conservative and expensive by planning for absolute worst-case scenarios.

The Multi-Energy Microgrid Challenge

The MEMG architecture integrates diverse components, including wind and solar units, battery storage, electrolyzers, and fuel cells. Coordinating these elements requires a dispatch strategy that can handle significant supply and demand uncertainty without sacrificing economic efficiency or system reliability.

Multi-Energy Microgrid (MEMG) Architecture

Proposed Methodology: Distributionally Robust Optimization

This paper proposes a two-stage Distributionally Robust Optimization (DRO) framework. Unlike standard approaches, DRO seeks the worst-case expected cost within a "Wasserstein ambiguity set"—a collection of potential distributions centered around empirical data. This provides a data-driven "risk knob" (the Wasserstein radius $r$) to tune the level of robustness between the optimism of SP and the conservatism of RO.

Operational Innovations

A significant contribution to the model's flexibility is the inclusion of a Three-State Electrolyzer Model. Traditional models often simplify electrolyzers to binary "on/off" states. By incorporating a "standby" state, the model captures the reduced costs and faster response times associated with keeping a unit warm versus a cold start, providing more agility during intra-day re-dispatch.

Three-State Electrolyzer Operational Model

The C&CG-DRO(CG) Algorithm

To overcome the computational hurdles of two-stage DRO, the authors developed a nested decomposition algorithm. The outer loop utilizes Column-and-Constraint Generation (C&CG), while the inner loop employs Column Generation (CG) to solve the worst-case expectation problem from a primal perspective. This structure is highly efficient and supports parallel computing, addressing the real-time needs of engineering applications.

Key Results

The proposed framework was tested on a simulated industrial MEMG with the following outcomes:

  • Reliability: The DRO approach significantly outperformed SP in out-of-sample tests, reducing electricity load shedding from 8.7% to 0.3% and heat load shedding from 12.1% to 1.2%.
  • Cost-Effectiveness: DRO achieved a more profitable dispatch than RO by avoiding excessive conservatism while maintaining high reliability.
  • Computational Performance: The C&CG-DRO(CG) algorithm provided a 10x speedup over standard implementations, solving complex cases within 4,000 seconds where traditional Benders-dual methods failed to find high-quality solutions within a 7,200-second limit.

Conclusion

The data-driven DRO framework offers a robust yet flexible solution for MEMG dispatch. By leveraging the Wasserstein metric and a novel primal-based decomposition algorithm, the model achieves high reliability and computational efficiency. Future work will explore multi-microgrid clusters and more complex network constraints, such as gas or heat pipe dynamics.

Summary updated

Abstract

This paper studies adaptive distributionally robust dispatch (DRD) of the multi-energy microgrid under supply and demand uncertainties. A Wasserstein ambiguity set is constructed to support data-driven decision-making. By fully leveraging the special structure of worst-case expectation from the primal perspective, a novel and high-efficient decomposition algorithm under the framework of column-and-constraint generation is customized and developed to address the computational burden. Numerical studies demonstrate the effectiveness of our DRD approach, and shed light on the interrelationship of it with the traditional dispatch approaches through stochastic programming and robust optimization schemes. Also, comparisons with popular algorithms in the literature for two-stage distributionally robust optimization verify the powerful capacity of our algorithm in computing the DRD problem.
BibTeX
@misc{sun2025data,
  author = {Xunhang Sun and Xiaoyu Cao and Bo Zeng and Miaomiao Li and Xiaohong Guan and Tamer Basar},
  title = {{Data-Driven Two-Stage Distributionally Robust Dispatch of Multi-Energy Microgrid}},
  year = {2025},
  eprint = {2504.09638},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2504.09638},
}
Figure