Estimação de Componentes Harmônicas em Sistemas Elétricos Utilizando Combinação Convexa de Filtros Adaptativos

This paper proposes the CONV-TNLMS architecture for harmonic component estimation in electrical power systems, based on the convex combination of scalar NLMS and tensor NLMS adaptive filters operating in parallel. Tensor modeling via Kronecker product decomposition reduces the number of adapted parameters, while the convex combination mechanism with recursive power normalization eliminates the trade-off between convergence speed and steady-state error. Experimental tests with real voltage data confirmed a THD of 9.93 % and an MSE floor below -20 dB, demonstrating the superiority of CVX-TNLMS over the classical CVX-NLMS approach.