HyperTran: High-Performance Transient Stability Simulation of Power Systems on Modern Parallel Computing Hardware

计算机仿真是预测电力系统大扰动后暂态稳定的一种广泛使用但计算量大的方法。目前，仿真性能落后于行业对在线、实时应用的需求和对数据驱动应用的研究需求，特别是对于可再生能源渗透率高的实际规模的系统。该NSF项目旨在开发一个高性能框架，使暂态稳定仿真的数据、任务和作业并行性能够扩展到当前和未来并行计算硬件的全部能力。提出的框架将为理解电力系统模型应该如何表示以及如何构建计算工作流以利用现代并行计算机带来革命性的变化。该项目的智力优势包括：a)通过设计电力系统模型的并行软件表示，加快了微分-代数方程(DAE)的建立和求解阶段；b)根据仿真测试用例的特点识别和利用计算方法和硬件设备。该项目的更广泛影响包括通过开源软件和出版物传播研究成果，整合研究和教育活动，以及增强电网基础设施稳定性的潜力。已确定三项任务来实现这一目标。任务1将创建电源模型和计算工作流的软件表示，以便为CPU和图形处理单元(GPU)上的DAE构建过程实现分段数据和任务并行。它将通过协调在优化缓存的同时更新跨模型共享的方程来确保并发执行的正确结果。任务2将开发自适应调度器，以确定和应用给定电力系统情况下最有效的硬件和解决方案算法，并考虑系统规模、加速技术和实际约束。任务3将研究用于在异类硬件上并行多场景作业的流水线算法，以构建和求解DAE以最大化硬件利用率。该项目成功完成后，预计将为现代计算硬件建立一个新颖、高性能的框架，显著加快电力系统动态模拟的速度。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Computer simulation is a widely used but computationally intensive method for predicting the transient stability of power systems following large disturbances. Currently, simulation performance lags behind industry demand for online, real-time applications and research need for data-driven applications, particularly for practically sized systems with high penetration of renewable energy. This NSF project aims to develop a high-performance framework that enables data, task, and job parallelisms for transient stability simulations to scale to the full capacity of contemporary and future parallel computing hardware. The proposed framework will bring transformative changes to the understanding of how power system models should be represented and how computational workflows should be structured to take advantage of modern parallel computers. The intellectual merits of the project include a) accelerating the building and solving phases of differential-algebraic equations (DAE) through the design of parallel-enabled software representations of power system models, and b) the identification and utilization of computational methods and hardware devices based on the characteristics of simulation test cases. The broader impacts of the project include the dissemination of research findings via open-source software and publications, integrated research and education activities, and the potential to enhance the stability of the power grid infrastructure.Three tasks have been identified to accomplish the goal. Task 1 will create software representations of power models and computational workflows to enable staged data and task parallelisms for the DAE building process on CPUs and Graphics Processing Units (GPUs). It will ensure correct results from concurrent executions by coordinating the updating of equations shared across models while optimizing caches. Task 2 will develop adaptive dispatchers to identify and apply the most efficient hardware and solution algorithms for given power system cases, considering system size, acceleration techniques, and practical constraints. Task 3 will investigate pipelining algorithms for parallelizing multi-scenario jobs on heterogeneous hardware to build and solve DAEs for maximized hardware utilization. Upon successful completion, the project is expected to have established a novel, high-performance framework for modern computing hardware that will markedly accelerate the simulation of power system dynamics.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.