Multilayer Perceptron Artificial Neural Network-Based Solution for Interharmonics Diagnosis in Frequency Spectra

Interharmonics diagnosis is increasingly important due to the rise of power electronic devices, renewable energy systems, and related technologies. The IEC 61000-4-7 standardizes the calculation of interharmonic amplitudes in a frequency spectrum; however, the standard has gaps in interharmonic identification and frequency estimation, which has led to alternative solutions in the literature that do not adhere to the framework established by the IEC 61000-4-7. This study proposes a novel interharmonics diagnosis solution based on Multilayer Perceptron Artificial Neural Networks. As a post-processing technique, the solution adopts the interharmonic group defined in IEC 61000-4-7 and complements the standardized method by verifying the existence of interharmonics in a spectrum of a measured, noisy signal, and estimating the frequency of the identified interharmonics. Validation of the proposed solution with synthetic signals presented a hit rate of 100% for interharmonics identification and an average relative error of 1.17% for the frequency estimation. Experimental results from a small power generation system with a variable speed drive are also evaluated. This new method, therefore, offers potential for interharmonics diagnosis in modern power grids.