Neural networks for phase and symmetrical components estimation in power systems

An original three-phase neural approach for phase and symmetrical components estimation is proposed in this paper. This neural structure can be used for power quality control, in an active power filtering scheme for example. The approach is composed of a neural symmetrical voltage components extraction and a neural phase detection technique. These functional tasks are decomposed and approximated by the learning of Adaline neural networks. The whole neural architecture is implemented on a digital signal processor (DSP) and is applied to a three-phase power system. The presented analysis substantiates the immunity of the proposed approach under distorted utility conditions. Experimental results, under various utility conditions, demonstrate its phase tracking and voltage components extraction abilities. Its performance and robustness are also compared to a conventional phase-lock loop (PLL) on a three-phase power supply with varying parameters and distorted by harmonics and random noise.