A real-time neuro-computing three-dimensional space vector algorithm for three-phase four-leg converters

Four-leg voltage source converters have successfully been used to nullify the zero-sequence current generated by unbalanced or nonlinear loads. This paper introduces an on-line, simple, intelligent, and computationally efficient neuro-computing classification algorithm for the implementation of three-dimensional space vector modulation (SVM) on four-leg voltage-source inverters. The proposed technique uses the concepts of counter propagation neural networks (CPN) for prism identification, and employs a nonlinear classifier network for tetrahedron identification. Nonlinear function approximations and bulky look up tables are successfully avoided, and exact positioning of the switching instants is obtained. Analytical analysis and simulations on a four-leg voltage-source converter validate the proposed scheme