Split-as-a-Pro：基于算子分割与交替投影的行为控制
摘要
本文介绍了 Split-as-a-Pro，一种将行为系统理论、算子分割方法和交替投影算法集成的控制框架。该框架将 arising in both control and estimation 的动力学优化问题简化为高效的投影计算。Split-as-a-Pro 基于一种非参数表述，利用系统结构将由各子系统施加的动力学约束与外部约束（如互联约束和输入/输出约束）分离。これにより、associated projection が高効率的計算可能であれば、任意のシステム表現形式を使用でき、スケーラビリティとグレイボックスモデリングの互換性が向上する。We demonstrate the effectiveness of Split-as-a-Pro by developing a distributed algorithm for solving finite-horizon linear quadratic control problems and illustrate its use in predictive control.当然，本研究通过为解决有限时域线性二次控制问题开发分布式算法，展示了Split-as-a-Pro的有效性，并说明了其在预测控制中的应用。Our numerical case studies show that algorithms obtained using Split-as-a-Pro significantly outperform their centralized counterparts in runtime and scalability across various standard graph topologies, while seamlessly leveraging both model-based and data-driven system representations.我们的數值案例研究表明，使用Split-as-a-Pro获得的算法在各种标准图拓扑结构下，在运行时间和可扩展性方面显著优于集中式方法，同时无缝地利用模型基础和数据驱动的系统表示。
引用
@article{arxiv.2505.19411,
title = {Split-as-a-Pro: behavioral control via operator splitting and alternating projections},
author = {Yu Tang and Carlo Cenedese and Alessio Rimoldi and Florian Dórfler and John Lygeros and Alberto Padoan},
journal= {arXiv preprint arXiv:2505.19411},
year = {2025}
}