Global Voltage Control in Medium-Voltage Grids Using On-Load Tap-Changers and Artificial Neural Networks

Abstract:
The increasing integration of distributed energy resources in medium-voltage networks leads to bidirectional power flows and spatially-temporally heterogeneous voltage profiles that challenge the ability to maintain voltage profiles within acceptable ranges in the whole grid. This paper develops a novel twostage neural control concept for global voltage optimization with limited measurement infrastructure. The developed system combines a regression model for voltage prediction at grid nodes with a downstream classification model for optimal OLTC tap switching. The extended neural control achieves a 54% improvement in global voltage deviation compared to the conventional OLTC control based on a user-defined voltage setpoint. The work confirms that neural networks can achieve global voltage optimization in medium-voltage network systems through datadriven feature extension, significantly outperforming conventional local methods. The proposed approach was validated using time-domain co-simulation.