Adversarial Learning as a PMU Signal Filtering Technique in Load Models Identification

A current trend in the area of load modeling in medium voltage busbars of an electric power system is to take advantage of ambient data (with small variations) recorded by Phasor Measurement Units (PMU) to estimate the parameters of the load models using automatic and online methodologies. When working with ambient PMU data, noise has a significant and negative impact on the parametric estimation of the load models. This is where signal filtering algorithms play a key role. In the literature lacks automatic and robust filtering algorithms designed for load modeling, especially with ambient PMU data. In order to solve this problem, this work proposes a filtering algorithm based on Artificial Neural Networks (ANN) and Adversarial Learning. The advantage of the proposed algorithm is that it does not need a design parameter to be adjusted and, therefore, it is ideal to work with different PMU measurements, either voltage, current, active power, or reactive power, indistinctly. Finally, the performance of the proposed algorithm is compared in two ways with respect to traditional filtering techniques recommended in the literature. On the one hand, based on the amount of noise removed and, on the other hand, on the accuracy achieved when the load model’s parameters are estimated (the latter being an aspect not previously explored in literature). It is concluded that the proposed algorithm achieves significantly better performance, especially with ambient PMU data.