A Fast and Universal Neuro-Based SVM Algorithm for Multi-Level Converters

This paper proposes a novel, simple and fast classification algorithm for implementation of Space Vector Modulation (SVM) method for a Diode Clamped Multi-level Converter (DCMC) with an arbitrary number of levels. The proposed algorithm is based on a classifier neural network which provides a straightforward and computationally efficient approach without the use of trigonometric calculations or look-up tables to identify (i) the location of reference voltage vector, (ii) its adjacent switching voltage vectors, and (iii) their corresponding on-duration time intervals. The feasibility of the proposed SVM algorithm is validated based on theoretical analysis, simulation studies and experimental tests on a DSP-controlled, 5 kVA, three-level converter system.