Adaline for Online Symmetrical Components and Phase-Angles Identification in Transmission Lines

This paper presents a new method for online symmetrical components and phase-angle extraction from high-voltage transmission-line faults. This method is based on the Adaline neural networks and the instantaneous power theory, also known as the p - q method. A new current decomposition is proposed in order to derive the direct, inverse, and homopolar current components. The average and oscillating terms of powers in the alpha beta frame are separated by using four Adaline neural networks. The Adalines use a cosine and sine as inputs in order to learn the linear combination of the powers. The resulting symmetrical components are used by three other Adalines for phase-angle estimation between direct and inverse current components. These phase angles permit classifying the fault types. The neural networks use an online learning process-based Widrow-Hoff algorithm and can adapt their weight parameters to the power-supply evolution. Simulation results show the performance and the robustness of this method and provide a perspective for protection relay improvement.