Smart Partitioning Based Large-Scale Power System Analysis on High-Performance Computing Platform: Modeling, Algorithms, and Computations

随着可再生能源、分布式发电、储能、智能负荷和新的市场驱动激励机制的不断整合，当今的电力系统正在经历一场朝着更随机和动态的电力系统运行范式的演变。为了处理由于电力系统的扩展和演变而导致的日益增长的规模、复杂性和异质性的数学问题，并且特别是借助于先进的计算技术来进行快速和准确的电力系统分析，项目组设想了一种新的“智能分区”概念，作为一种创新的分解方法，以加强对大型通过利用高性能并行计算平台来扩展复杂的电力系统。该研究将系统地探索这一新的区域分解概念的问题建模，解决方案的算法和计算实现，以改变如何在高性能并行计算平台上有效地进行稳态和动态大规模电力系统分析。该研究将有助于加快基于高性能并行计算平台的应用程序的扩散，以提高电力系统的经济性，可靠性和稳定性，并加速其他智能电网技术的开发和部署。作为该项目的一个组成部分，教育计划的重点是高性能并行计算技术在电力系统中的应用，并强调如何将研究中获得的知识和成果通过标准的学术手段直接导入项目的教育目标，激发学生对大规模电力系统研究的兴趣，所提出的基于智能划分的分解方法通过构建基于虚拟域的、灵活的、可扩展的和高效的划分结构，并利用并行优化科学和并行计算技术来驯服大规模电力系统分析问题的规模和计算难度。拟议的电力系统分解将不再需要围绕一个特定的域，如功能，场景，地理区域或时间制定。相反，它可以基于一个或一组虚拟域来充分利用电力系统的固有结构、物理一致性、组件特性和独特属性。所提出的智能分区策略具有与现代电力系统分析的目标相一致的独特属性，包括建模的灵活性和可扩展性，算法的可扩展性和并行性，以及计算的效率和可用性。主要研究任务包括：（1）开发基于虚拟区域分解的电力系统稳态和动态柔性建模方法;（2）探索可扩展的、有效的并行计算算法，以完全并行的方式处理电力系统研究，优化计算任务之间的信息交互;以及（3）实现高效和快速的高性能并行计算，其最佳地将计算任务映射到处理器并平衡任务的布置。

With the ongoing integration of renewable energy, distributed generation, energy storage, smart loads, and new market-driven incentives, today's power system is undergoing an evolutionary transformation towards a more stochastic and dynamic paradigm for power system operation. In order to handle the ever-growing size, complexity, and heterogeneity of the mathematical problems resulting from power system expansion and evolution, and to conduct rapid and accurate power system analyses especially with the aid of advanced computing techniques, the project team envisions a new concept of "Smart Partitioning" as an innovative decomposition method to enhance the analysis of large-scale and complex power systems by leveraging a high performance parallel computing platform. The research will systematically explore this novel domain decomposition concept in terms of problem modeling, solution algorithms and computational implementation to transform how steady state and dynamic large-scale power system analyses are efficiently performed on a high performance parallel computing platform. The research will help expedite the proliferation of applications based on the high-performance parallel computing platform to enhance the economics, reliability and stability of electric power systems and accelerate the development and deployment of other smart grid technologies. As an integral part of this project, the educational plan focuses on power system applications on high-performance parallel computing technology and emphasizes how the knowledge and results gained from the research can be directly channeled into the project's education goals via standard academic means, stimulate students' interest in large-scale power system studies, and empower industry engineers with skills for life-long learning.The proposed smart partitioning based decomposition method could revolutionize power system analysis by constructing a virtual domain based, flexible, scalable, and efficient partitioning structure, and utilizing parallel optimization science and parallel computing techniques to tame the size and computational difficulty of large-scale power system analysis problems. The proposed power system decomposition will no longer need to be formulated around one particular domain like functions, scenarios, geographical areas or time. Instead, it can be made based on one or a set of virtual domains to fully exploit the inherent structures, physical coherency, component characteristics and unique properties of the power system. The proposed smart partition strategies have distinct and unique attributes that are consistent with the objectives of modern power system analysis, including flexibility and extensibility of the modeling, scalability and parallelism of the algorithms, and efficiency and usability of the computations. The key research tasks include (1) developing virtual domain decomposition based flexible modeling for both steady state and dynamic analyses of the power system composed of heterogeneous types of power production and delivery elements; (2) exploring scalable and effective parallel computing algorithms that process the proposed power system studies in a fully parallel manner and optimize the information interactions among calculation tasks; and (3) implementing efficient and rapid high-performance parallel computations that optimally map the computation tasks to processors and balance the arrangement of tasks.