Transactions of China Electrotechnical Society  2023, Vol. 38 Issue (16): 4246-4262    DOI: 10.19595/j.cnki.1000-6753.tces.220982
Current Issue| Next Issue| Archive| Adv Search |
Numerical Model Decoupling Acceleration Method with Numerical Integration Parallelism Construction Based on Computation Front in Real-Time Simulation
He Shaomin1, Zhang Zhe1, Lu Yiping1, Yang Huan1, Shen Jie2
1. College of Electrical Engineering Zhejiang University Hangzhou 310027 China;
2. Leadrive Technology (Shanghai) Co. Ltd Shanghai 201315 China

Download: PDF (5711 KB)   HTML (1 KB) 
Export: BibTeX | EndNote (RIS)      
Abstract  The computation speed is the key performance of numerical simulation, which is the foundation and underlying core technology of intelligent computing. However, with the application of high switching- frequency power devices in electric drive systems, the constraint of short computation time has led to great challenges for computation speed in real-time or online numerical simulation. At the same time, with the trend of parallel architecture in digital platforms, parallel computing is considered an effective way to optimize the computation speed with short time constraints. However, the state-of-the-art numerical models and algorithms are based on serial computation, lacking of redundant parallelism. Therefore, the computation front design methodology is proposed to construct the redundant parallelism for numerical integration, so the numerical model can be decoupled and parallel computed based on the redundant parallelism to accelerate the numerical simulation.
Firstly, the relationship between numerical integration and algorithm structure in power electronic simulation is discussed, which is the key to the numerical solution of ordinary differential equations (ODE). The computation front involved in numerical integration optimization is proposed, and the redundant parallelism construction process of the optimization method is described based on the predictor-corrector method. Secondly, the influence of the proposed method on the numerical performance is evaluated from numerical precision and numerical stability. The parallel computing resources are utilized in the proposed method to improve the numerical precision without increasing the computation time. Although the numerical stability domain changes during the optimization process, the influence on numerical simulation is limited, especially in applications with small step sizes. Thirdly, the possibility of general application is discussed from the multistep-step integration method and high-order integration method, proving that the computation front can be extended to different integration methods. Finally, the parallel numerical model is designed and applied to the electrical drive system based on the redundant parallelism constructed by the proposed method. As a result, the model-level numerical acceleration is achieved.
The real-time simulation environment of hardware in the loop (HIL) is demonstrated to verify the proposed method. The real-time simulation with a serial computing model is compared with offline simulation in Simulink, so the effectiveness of self-defined HIL based on NI PXI is verified. Then, the real-time simulation with a parallel computing model is compared between the latency insertion method and the proposed method. The results show that the relative error with the proposed method is reduced to less than 3 %, and the numerical precision of parallel simulation is improved. Moreover, the speedup ratio and parallel efficiency are 1.72 and 57.2 %, respectively, which expands the step size selection range of real-time simulation.
The following conclusions can be drawn from the real-time simulation analysis: (1) The redundant parallelism of numerical integration can be constructed based on the computation front to realize parallel computation. (2) The local step size of numerical integration is potentially changed during optimization, which may affect the numerical performance. Thus, the proposed method with accuracy compensation is designed based on the predictor-corrector method, which can reduce the simulation error. (3) The parallel acceleration can be achieved without accuracy reduction, but the parallel efficiency is limited due to an unbalanced computation load, which could be discussed in future research.
Key wordsParallel computation      electrical drive system      computation front      real-time simulation      numerical acceleration      numerical integration     
Received: 31 May 2022     
PACS: TM46  
  TP337  
Service
E-mail this article
Add to my bookshelf
Add to citation manager
E-mail Alert
RSS
Articles by authors
He Shaomin
Zhang Zhe
Lu Yiping
Yang Huan
Shen Jie
Cite this article:   
He Shaomin,Zhang Zhe,Lu Yiping等. Numerical Model Decoupling Acceleration Method with Numerical Integration Parallelism Construction Based on Computation Front in Real-Time Simulation[J]. Transactions of China Electrotechnical Society, 2023, 38(16): 4246-4262.
URL:  
https://dgjsxb.ces-transaction.com/EN/10.19595/j.cnki.1000-6753.tces.220982     OR     https://dgjsxb.ces-transaction.com/EN/Y2023/V38/I16/4246
Copyright © Transactions of China Electrotechnical Society
Supported by: Beijing Magtech