Frequency Domain Relaxation of Power Systems

在电力系统的分析与仿真中，暂态稳定问题是计算量最大的问题之一。因此，在这个领域，高速并行和矢量计算机今天可用，将发现在近期和遥远的将来最适用。提高数字计算引擎计算能力的一个当前趋势是使用廉价的中速处理器和中到大规模并行（10-1000处理器），如sequentbalance、Encore Multimax、Intel iPSC、BBN Butterfly。另一个趋势是使用更昂贵的高速矢量处理器和小规模并行（1-10），如CRAY和IBM矢量处理器的并行管道，而其他计算机在这两个阵营中都有涉足。比如Alliant。基于频域的算法，而不是传统的时域算法，显示出三个重要的特征，这些特征有可能在上述机器中利用专业架构。首先，频域技术通常导致算法非常自然地可并行化，而时域算法受到数值积分方案固有的先例关系的影响，并限制了可利用的并行性的数量。其次，将频域技术应用于电力系统动力学问题，通常会产生富含高度可向积计算的算法，与时域算法形成鲜明对比，后者的代码向量化非常有限。第三，可以很容易地利用频域数据压缩。

It is well recognized that one of the most computationally intensive problems encountered in the analysis and simulation of electric power systems if the transient stability problem. Hence it is this area in which the high speed parallel and vector computers being made available today, will find the most applicability in the near and distant future. One current trend in increasing the computational power of digital computing engines is to use inexpensive moderate speed processors and medium to large scale parallelism (10-1000 processors) such as the Sequent Balance, Encore Multimax, Intel iPSC, BBN Butterfly. Another trend is to use the more expensive high speed vector processors and small scale parallelism (1-10) such as the CRAY and IBM vector processors with parallel pipelines, while there are other computers with a foot in both camps, such as the Alliant. Frequency domain based algorithms, rather than traditional time domain algorithms, display three important characteristics which have the potential for utilizing the specialist architectures in the machines above. First, frequency domain techniques often lead to algorithms which are very naturally parallelizable, whereas time domain algorithms suffer from precedent relationships which are inherent in numerical integration schemes and limit the amount of exploitable parallelism. Second, frequency domain techniques applied to power system dynamics problems, often result in algorithms which are rich in highly vectorizable inner product calculations, contrasting the time domain algorithms, in which code vectorization is extremely limited. Third, frequency domain data compression can be easily exploited.