Neural adaptive notch filter to harmonic detection

Abstract

This paper presents the application of a neural adaptive notch filter for harmonic detection. The final objective of the strategy is improving the speed of convergence of the Neural Adaptive Filter and reduces the steady-state error. The Recursive Least Square (RLS) algorithm was used to adjust the coefficients. The new strategy proposed doesn't use a Phased Looked Loop (PLL) to generate the orthogonal input signals for the neural adaptive notch filter. These two signals are generated with Clarke transformation of the load current and then the two orthogonal signals are filtered. With this, the total transient response drops below one cycle of the fundamental frequency (60Hz). Simulations using Matlab/Simulink are presented to clarify the algorithm. Also practical implementation is performed using the DSP Texas Instruments TMS320F2812 in a shunt active power filters (SAPF) and the results depicted.