Robust Tuning of Cascaded PI Controllers for Three-Phase Vienna Rectifiers Using Multi-Scenario Min-Max Cost

Authors

  • Mohammed Obayes Yousif Al-Furat Al-Awsat Technical University, Al-Musayyib Polytechnic College, Iraq.

DOI:

https://doi.org/10.17977/um048v8.i1.2026.33-44

Keywords:

Vienna rectifier, Cascaded PI Control, Genetic Algorithm, Robust Tuning, Min-Max Optimization

Abstract

The three-phase Vienna rectifier is commonly used for power factor correction in various power systems. Its functionality depends heavily on gain values in the cascaded PI controllers used in the current, DC bus voltage and neutral point voltage control loops. Commonly, heuristic algorithms are used to tune these gains; however, they are often optimized for a specific operating point. To overcome this issue, this paper proposes a multi scenario tuning process whereby gain values are tested under nine different operating scenarios encompassing; light and rated loading, loading transients, low grid strength, grid imbalance and component mismatch (inductance and capacitance). A genetic algorithm (GA) is used to reduce the min-max cost function. This is compared to a traditional manually designed controller (20Hz voltage loop bandwidth), a quick manually designed controller (50Hz), a single scenario tuned GA and a particle swarm optimiser using the same robust cost function. Each controller was tested on twelve scenarios, three of which were unseen by the optimizers. The averaged and PWM switched models were used in MATLAB. The single scenario GA controller produces the best results around its operating point but fails in the most challenging scenario producing a total harmonic distortion (THD) of up to 14.5% (averaged model) and 21.3% (switched model) whilst having a power factor of 0.78. The proposed robust GA model can maintain the worst case THD of the current to only 2.56% compared to 4.12% and 4.46% for the manually designed controllers whilst maintaining a power factor of above 0.9996. It also reduces the worst case DC link voltage deviation when loading changes to only 4.3V compared to 14.0V and a reduced settling time of 18.3ms compared to 75.9ms. The robust GA model and robust particle swarm optimization algorithm produce nearly identical results

Downloads

Published

2026-11-08

Issue

Section

Articles