Distortions Identification With Artificial Neural Networks Based on Symmetrical Components

A new neural approach for identifying and compensating for harmonic distortions is proposed in this paper. Based on a new signal decomposition of the voltage and the current, this method identifies the direct, inverse and homopolar components of both the voltages and the currents of the supply network. For one signal, the component extraction requires four independant Adaline networks. The components are then used to on-line compute the instantaneous direct, inverse and homopolar powers. The learning capabilities of the Adalines allow to determine the APF (Active Power Filter) compensation currents under fluctuating conditions and according to the compensation objectives, i.e., harmonic compensation, power factor correction, or selective harmonic compensation. The resulting compensation currents are re-injected phase-opposite through an APF in real-time. The performance is evaluated through several simulation examples. The results show that the proposed neural method outperforms other methods, such as the conventional instantaneous power theory.