Control strategies for active filters using pattern recognition techniques

In recent years there has been considerable interest in the use of active filters for eliminating harmonic currents in power supply systems. Most filter configurations require detection of the harmonic content and generating switching waveforms for the filter. The feasibility of using pattern recognition techniques for detection of harmonics and generation of information suitable for controlling active filters is investigated. The relative merits of time and frequency domain representation are studied. The experimental investigation uses fast Fourier transforms to detect the harmonic spectra which are used to train the system and store the corresponding pre-determined switching patterns for control of a filter. On receiving an unknown signal the system compares it with the stored image and identifies the appropriate switching pattern for the filter. The feasibility of this approach is assessed in an application incorporating a switched-capacitor filter to eliminate current harmonics in the input current to a rectifier system such as that used for uninterruptible power supply (UPS) systems