GPU Parallel-Rate Exponential Integrator Algorithm for Efficient Simulation of Power Electronic Systems

Electromagnetic transient (EMT) simulation of power electronic converters is critical for analysis, design, and fast control prototyping of power and energy systems. This paper proposes a multi-granular GPU parallel-rate exponential integrator (EI) algorithm for fast offline EMT simulation of power electronic systems. The proposed parallel-rate EI algorithm utilizes the massively parallel GPU architecture to compute multiple discretization steps in parallel. The matrix-vector computations of the EI algorithm within each time step are also parallelized. Additionally, a novel GPU-based framework is proposed for numerically efficient precomputation of matrix exponentials before a simulation loop starts. The high degree of parallelism leads to large simulation speedups compared to single-thread CPU implementations. The discretization technique of high-order EI algorithm is absolutely stable with no numerical ringing and can achieve accurate differential equation discretization with large time-step sizes. The proposed parallel-rate EI solver is further applied to detect passive/diode switching events accurately. Two case studies are used to demonstrate the accuracy and efficiency of the parallel-rate EI algorithm. Two additional case studies showcase the benefit of the proposed parallel precomputation technique for matrix exponentials.