Improving Monte Carlo Analysis of Power Systems With Data Clustering and Vector Fitting

Both hierarchical agglomerative clustering (HAC) and vector fitting (VF) can be used to make Monte Carlo analysis of power systems faster to perform and more informative. HAC is used to identify power system frequency responses as sufficiently similar to one another. VF is used to produce transfer functions approximating impedance characteristics. This paper describes a mix of existing and new techniques that reduce the number of cases, reduce the order of the system, and provide insight as to what parameters are most important. To demonstrate these techniques, a harmonic emission study for a large industrial facility is performed. The techniques reduce an 8.7-year computation of 1.68 million cases to an 80-hour computation of 2700 cases while maintaining a user-specified accuracy bound.