Distortion Compensation and Power Factor Improvement through Three-Phase Switched Capacitor Active Power Filter (SCAPF) with Neural Network and Band-Less Hysteresis
By: PROF. DR. NASRUDIN BIN ABD RAHIM
Increased non-linear loads in electricity supply networks produce non-sinusoidal current distorted by harmonics (which must be compensated) and cause low line-power-factor (which must be corrected). The proposed combination of neural network and band-less digital hysteresis switching algorithm is a controller, forcing the supply current to be sinusoidal, have low current harmonics, and be in phase with the line voltage.
Among harmonic-mitigating topologies, active power filter (APF) is the most rugged. SCAPF topology brings a new dimension to APF, as at low switching frequencies less components (especially capacitors) and low ratings are possible. This work proposes a topology for three-phase applications, continuing from a previous R&D on single-phase-application SCAPF .
This work also focuses on just two parts of the control techniques for APFs: reference generation (known also as harmonic detection) and current control (known also as switching strategy), simplifying the system and easily implementing the control algorithm in a small, limited-speed, limited-memory, embedded controller.
Extraction of harmonics poses various problems. In the single-phase SCAPF, computation/convergence of its off-line control optimisation algorithm took time, slowing down response to harmonic current variations. Genetic algorithm, developed later, reduced convergence time but was no less complex or easier to implement, hardware-wise or as on-line control. Hysteresis current control, principled on adaptive noise cancelling, needs both analog and logic circuits, complicating hardware development. This work proposes a fast and improved artificial neural network control scheme, for its simplicity and learning and generalisation ability, also for possible control scheme modification through a weights-updating algorithm that hastens harmonics extraction, adaptation, and the APF’s response.
*Results(before filtering): Simulated waveforms for the source voltage VS, the load current IL, the filter current IF, and the source current IS for capacitive load.
Result(after filtering): Simulated waveforms for the source voltage VS, the load current IL, the filter current IF, and the source current IS for inductive load.
(The original manuscript of this research paper was accepted 17 December 2008 by IEEE Transactions on Power Electronics.)
 M.A.M. Radzi and N.A. Rahim, “Neural Network and Band-Less Hysteresis Approach to Control Switched Capacitor Active Power Filter for Reduction of Harmonics,” IEEE Trans. Ind. Electron.
The researchers, Mohd. Amran Mohd. Radzi & Prof. Dr. Nasrudin Abd. Rahim, are with the University of Malaya’s Centre of Research for Power Electronics, Drives, Automation and Control (UMPEDAC).
For more information on the research, Please contact: PROF. DR. NASRUDIN BIN ABD RAHIM