Adaline neural networks for online extracting the direct, inverse and homopolar voltage components from a composite voltage

This work describes an improved Adaline neural networks method for online extracting the direct, inverse and homopolar voltage components from a composite voltage. A new voltage decomposition is thus proposed and developed. These skills are transferred to four Adalines by fixing their inputs. Adaline neural networks are used with a LMS learning process to compute the weights biases and thus to find out the amplitude and the phase of the direct, inverse and homopolar voltages of the electrical network. The learning allows an online adaptation to the changing parameters of the electrical network, e.g., nonlinear and time-varying loads. Compliant reference signals, i.e., sinusoidal and equilibrated voltages, are obtained from the proposed neural method, enhancing the harmonics compensations performance. A comparison with a conventional PLL is also addressed. Simulations and experiments are reported