Effective Parallelism for Equation and Jacobian Evaluation in Power Flow Calculation
Abstract : This letter investigates parallelism approaches for equation and Jacobian evaluations in power flow calculation. Two levels of parallelism are proposed and analyzed: inter-model parallelism, which evaluates models in parallel, and intra-model parallelism, which evaluates calculations within each model in parallel. Parallelism techniques such as multi-threading and single instruction multiple data (SIMD) vectorization are discussed, implemented, and benchmarked as six calculation workflows. Case studies on the 70,000-bus synthetic grid show that equation evaluations can be accelerated by ten times, and the overall Newton power flow advances the state of the art by 20%.
? In some cases, the speedup is measured in comparison to the existing commercial library, and in other cases, the performance is evaluated compared to the CPU alone.
? Since the speedup reference is differently represented in each paper, it is difficult to understand the inherent speedup value in each case; however, it is clear that GPUs aid speedup when performing PF computations.
? Recent studies applying GPUs to PF studies have usually been conducted based on existing parallel processing research.
? Therefore, in terms of PF studies, a review of current parallel processing research is necessary to understand the application of GPUs.
? There have been many attempts to develop parallel algorithms optimized for different parallel architectures for efficiently solving the power flow problem.
? Although parallel computers have been successful in solving several computation-intensive problems, their high prices and long design and development cycles, and the high cost of maintaining them often make their long term availability unpredictable.
? The repetitive solution of the linear equations in Newton’s method is very time-consuming for large networks, if the problem is solved sequentially.
• The GS method is often used because it is the first practical approach proposed for estimating power flow in large-scale power systems.
• In, the basics of GPU-based parallel computing were explicitly presented, and an efficient method of using GPUs was proposed in. There are some other attempts to GPU-accelerated SLS solution using LU decomposition in.
• The speed of their proposed solver was up to 76 times greater compared to the KLU library. KLU library is a solver for a sparse linear system.
• In, a GPU-based Chebyshev preconditioner integrated with a GPU-based conjugate gradient solver was proposed to solve SLS for speedup.
? The major advantages of MPoPCs include flexibility (supported by both embedded programmable logic and microprocessors) and low cost.
? Some research also has shown that SuperLU does not result in performance gains for circuit simulation matrices.
? FPGA-based configurable computing machines have shown significant results in the last decade for improving the performance of algorithms in numerous fields, such as DSP, data communication, genetics, image processing, pattern recognition, etc.
? With advances in recent years, the peak floating-point performance of FPGAs has outnumbered that of modern microprocessors and is growing much faster than the latter.
|