Measurement of Ground-Neutral Currents in Three Phase Transformers Using a Genetically Evolved Shaping Filter

Abstract

Measuring the current in the neutral-grounding resistor is needed for monitoring resistance-grounded three phase transformers. This current is limited to hundreds of amperes in case of a fault, and are almost negligible otherwise. The current transformer that senses the current must be rated for the fault conditions, thus it is difficult to obtain a precise measurement of the current when there is not a ground-fault in the system.

In this paper we propose a computer-based method for filtering the output of the current transformer and improving its accuracy for small currents. This processing is complicated, as the amount of noise is very high, and this noise is strongly correlated with the useful signal. We propose to use Kalman filtering, based on a model of the system, and augment the state of this model with a shaping filter, whose frequency response, when fed with white Gaussian noise, reproduces our measurements of the ambient noise. In particular, since the Power Spectral Density (PSD) of the noise changes with time, we propose to use a possibilistic description of the PSD of the noise, and search for a model whose PSD is between the soft margins defined by the possibilistic model. We will use a state-space based representation and a genetic algorithm, guided by a fuzzy fitness function, for evolving the shaping filter that best matches the ambient noise. The proposed method has been evaluated with field data captured at a 130KV substation transformer at La Corredoria (Asturias, Spain).