Parallel-in-Time-and-Space Data-Oriented Electromagnetic Transient and Dynamic Simulation of AC-DC Smart Grid

• Author / Creator

  Cheng, Tianshi

• This thesis pioneers new PiT and PiS acceleration techniques in the transient simulation of smart grids, introducing a series of algorithms, software technologies, and heterogeneous parallel computing practices that are the first of their kind in this field. It begins with an exploration of new PiT algorithms, particularly focusing on the application of the Parareal algorithm in accelerating AC-DC power grid simulations. The algorithm study addresses a series of theoretical and engineering challenges associated with PiT algorithms, demonstrating their value through the implementation of a massively parallel PiT+PiS algorithm on a heterogeneous CPU-GPU architecture for EMT-TS co-simulation scenarios.

  Meanwhile, the complexity of heterogeneous PiT+PiS algorithm implementation has brought new challenges to traditional software design philosophies. Further exploration revealed the advantages of new data-oriented software architectures over traditional object-oriented architecture to address the issues for exploring new applications and algorithms. In response, the first data-oriented cyber-physical power system simulation platform: ECS-Grid is developed based on the novel ECS architecture to address the demanding cyber-physical co-simulation issues for smart grids. The ECS-Grid provided better flexibility and performance compared to traditional solutions in complex cyber security scenarios running on distributed real-time hardware. An interdisciplinary digital-twin simulation solution leveraging LEO satellite networks is further developed to demonstrate the flexibility and superiority of the new data-oriented ECS-Grid platform. The studies have shown that the data-oriented ECS paradigm can be a promising choice for developing massively parallel heterogeneous PiT and PiS simulation programs of smart grids.
  
  To address the computational challenges posed by the massive integration of nonlinear models of renewable energy sources in smart grids, a new PiS acceleration method based on machine learning and neural network technologies is proposed. This novel method uses traditional nonlinear simulation models and Monte Carlo simulations to generate reliable training data for data-driven machine learning models. With the powerful ECS-based architecture, the optimal batching parallel processing on GPUs is achieved by modular designs, which fully leverage the technical advantages of the ECS-Grid platform. The proposed new machine-learning-reinforced parallel acceleration method has shown significant speed-up compared to traditional sequential nonlinear circuit simulation and can better utilize the heterogeneous hardware resource.
  
  Overall, the first-of-their-kind contributions of this thesis not only advance the theory and practical application of smart grid simulations but also pave the way for future innovations in power system transient analysis and applications.
  

• Subjects / Keywords

  • Parallel-in-Time
  • Data-Oriented
  • SmartGrid
  • EMT Simulation
  • Entity-Component-System

• Graduation date

  Fall 2024

• Type of Item

  Thesis

• Degree

  Doctor of Philosophy

• DOI

  https://doi.org/10.7939/r3-gh0g-hb93

• License

  This thesis is made available by the University of Alberta Library with permission of the copyright owner solely for non-commercial purposes. This thesis, or any portion thereof, may not otherwise be copied or reproduced without the written consent of the copyright owner, except to the extent permitted by Canadian copyright law.