AN IMPROVED METHOD OF BUSBAR VOLTAGE RECONSTRUCTION FROM SIGNALS OF ELECTRIC FIELD SENSORS INSTALLED IN AN INDOOR MV SUBSTATION

This paper presents an improved method for the reconstruction of busbar voltage waveforms from signals acquired by a system of electric field (EF) sensors located in an indoor medium voltage substation. In the previous work [8], the authors proposed the use of black-box models in the form of artificial neural networks (ANNs) for this task. In this paper it is shown that a parametric model of the system of EF sensors can reconstruct voltages with much lower errors, provided that it is accurately identified. The model identification is done by minimization of a nonlinear goal function, i.e. mean squared error (MSE) of voltage reconstruction. As a result of examining several optimization techniques, the method based on simulated annealing extended with a simplex search, is proposed. The performance of the model identified with this method is at least 8 times better in terms of MSE and at least 12 times better in terms of frequency domain errors than the best one of concurrent ANNs.