Optimizing basis function pole locations for transformer frequency response identification

In this paper a technique for obtaining accurate parametric models of wideband frequency domain systems is applied to modeling the frequency response of power transformers. Obtaining accurate frequency response models is a problem for conventional identification techniques, and typically results in ill-conditioning due to the frequency range and numerous resonant modes of the transformer. `Frequency Localising Basis Functions' improve the conditioning of the estimator, and hence provide a more accurate fit. The location of these functions are optimized using Particle Swarm Optimization, and applied to frequency response data taken from a power transformer. A case study is undertaken on a 132kV, 60MVA power transformer.