Artificial Neural Networks for Harmonic Estimationin Low-Voltage Power Systems

摘要

Harmonic estimation is the foundation of every activernnoise canceling method in low-voltage power systems.rnReference currents are generated and re-injected inrnphase opposition through an active power linernconditioner. Active Power Filters (APFs) are today thernmost widely used systems to compensate harmonics inrnindustrial power plants. We propose to improve thernperformances of conventional APFs by using artificialrnneural networks (ANNs) for harmonics estimation. Thisrnnew method combines both the advantages ofrnconventional APF to compute instantaneous real andrnimaginary powers and the learning capabilities of ANNsrnto adaptively choose the parameters of the powerrnsystem. In fact, the separation of the powers isrnimplemented with an Adaline neural network whichrnuses a priori known frequencies as inputs. Furthermore,rnmultilayer feedforward networks are used tornapproximate the instantaneous powers and to computernthe reference currents. Simulation results show thernreliability of the method and better performances thanrnconventional APFs.