Data-Driven Modeling of Nonlinear Transducers for Harmonic Measurements

Harmonic contents can heavily influence the performance of instrument transformers and sensors in the Power Grid, leading to measurement inaccuracies and potential failures. This paper presents a novel data-driven methodology for modeling non-linear transducers within the Power Quality (PQ) frequency domain. Traditional modeling approaches often depend on predefined analytical formulations or efforts to derive them, which can restrict adaptability to varying operating conditions. In contrast, this method leverages machine learning to eliminate the need for prior analytical knowledge. Central to this methodology is utilizing Artificial Neural Networks (ANN) to precisely characterize the relationships between harmonic components in input and output signals (I/O) exhibiting distinct harmonic patterns. The resulting data-driven model closely aligns with source data, preserving statistical intricacies observed in both simulated and real-world measurements, with deviations typically within a few parts in 10-5.