A neural network‐based selective harmonic elimination scheme for five‐level inverter. (6th September 2021)

• Record Type: Journal Article
• Authors: Toubal Maamar, Alla Eddine; Helaimi, M'hamed; Taleb, Rachid; Kermadi, Mostefa; Mekhilef, Saad
• Is Part Of: International journal of circuit theory and applications. Volume 50:Number 1(2022)
• Page Start: 298
• Page End: 316
• Publication Date: 2021-09-06
• Subjects: artificial neural networks (ANNs) -- C2000 Delfino (DSP) microcontroller -- five‐level inverter -- power electronics -- selective harmonic elimination (SHE); Electric circuit analysis -- Periodicals; 621.319205
• DOI: 10.1002/cta.3130
• Languages: English
• ISSNs: 0098-9886

Abstract

Abstract: This paper presents the implementation of selective harmonic elimination ( SHE ) in a five‐level inverter structure using artificial neural networks ( ANNs ). SHE is an effective low‐frequency modulation technique to eliminate selected harmonics and control multilevel converters. The use of ANN‐SHE requires the calculation of the optimum values of switching angles via the solving system of nonlinear equations for the total harmonic distortion ( THD ) reduction, where the nonlinear equations are founded by the complex Fourier series analysis of the inverter output voltage. The procured switching angle values are directly implemented by a multilayer perceptron ( MLP ) algorithm without a lookup table. The ANN model is obtained by training the neural network ( NN ), taking the modulation index ( M ) as an input and approximating switching angles as an output. A thorough analysis was carried out to show the programming steps of the proposed ANN‐based SHE using Matlab/Simulink environment. A realized inverter prototype steered by the proposed ANN‐based SHE was tested with various modulation indexes on a real‐time mode using a digital signal processor ( DSP ) C2000 Delfino–TMS320F28379D‐embedded board. A comparison between the simulation results and the experimental data is presented. The obtained results illustrate that the experimental results match the simulation closely, and the ANN model provides a fast and precise estimate of the switching angles for each value of the modulation index.